EY Generative AI

use cases repository

All sector use cases

June 2024

Written by
Brief guidance on use of deck

Use case category description1 Notes

 Meta data sheets
Use case category Description – These refer to next few pages. These are organized by sections, provides list of
Gen AI use cases by title and their application to various functions and sectors.

L4: Implemented  Has been (or being) implemented, with client  How sections are organized
with client  Can be a proof of concept, prototype, or solution

Each section
separators represent
 Has been (or being) built, for EY internal use one set of use cases
L3: Built for EY  Can be a proof of concept, prototype, or solution based on the source.
 Plan could be to also deploy with client

 Use case contact details:

 Has a demo or a solution available
L2: Built as a – Please refer slide notes or off-slide stickers for the use case contact details
 Built as an accelerator to speed up time to deliver, with
Prototype
client
 Use case application to different sectors
– Many use cases might apply to multiple sector. Sector tagging has been
provided on meta-data sheets or off-slide stickers.

L1: Proof of concept  Use case is still conceptual or has a proof of concept
Please note that the Gen AI use case repository is compilation of Gen AI use cases
from various sources. If you have new use cases from internal project or client
work and want to get those included in this central repository, or if you have any
update on the use case already included here, please feel free to reach out to us.

1.Please note that for now, some of the use cases have been provided with category description (on the top right corner of the use case slide).
On an ongoing basis , the use case category description would be updated or added if not already there.
 Generative AI use cases (from Section Sector Gen AI GTM decks)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)
Knowledge Management,
1 Integrating OpenAI into core programs Access, Analytics X X X X X X X X
Customer experience
2 LLMs to automate Funding Proposal Drafting Operations Customer Service X X X X X X X X
3 LLM model for ERP interface or ChatERP Operations Customer Service X X X X X X X X
4 LLMs for Knowledge Management Knowledge Management Customer Service X
Technology & Product Product
5 LLM for digitization of on-road rules X
development development
Category Management & Supplier
6 Supplier Contract Analyzer X X X X X
Supply Chain management
7 AI in Multifidelity Drug Design Research & development Screening X
Product design,
8 Generative AI-enabled product design Research & development X X
innovation
9 AI / ML Implementation IT & Technology Automation X X X X X X X X

10 RPA Center of Excellence IT & Technology Automation X X X X X X X X

11 Mixed Media Modelling Optimization Operations Optimization X X X X X X X

12 GenAI Asset management solution Operations Asset management X X X X X X X

SourceGPT – Bid response assistant for the deregulated
13 Supply Chain Procurement X X X X X X
markets
EcoPeer– An Industry ESG Supplier Risk Analysis and
14 Supply Chain Supplier risk X X X X X X
Benchmarking
15 GenAI-enabled product description Sales & Marketing Marketing X X

16 Product description generator Risk Management Risk Management X X

 Generative AI use cases (from Section Sector Gen AI GTM decks)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)

17 Product Listing Writer Sales & Marketing Marketing X X

18 Trend based curation via demand sensing Sales & Marketing Consumer insights X X X
Customer
19 Auto-generated landing page Sales & Marketing X
experience
Category Management & Contract
20 Use case combination – Fact-based negotiation X
Supply Chain management

21 EY Modern Finance Financial Management Corporate Analytics X X X X X X X X

Customer sales &
22 Wind Technician AI Assistant Sales & Marketing X
service

23 Sentiment search or Standard bookshelf Knowledge Management Search X X X X X X X X

Strategy & innovation,
24 AI Factory Concept: Single platform to deploy AI at scale Innovation X
Research & development

25 Staffing Model - Pod Structure IT & Technology AI Innovation X X X X X X X X

Payroll
26 GPT-4 powered chatbot for payroll inquiries Financial Management X X X X X X X X
Management
Customer Service - Blending virtual and human assistance
27 Technology AI Innovation X X X X X X X X
in customers support experiences
 Generative AI use cases (from Section GDS DNA Consulting team)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)
Generation,
Banker knowledge augmentation from videos Knowledge
0.1 summarization & X
Management
extraction
Supplier and Demand Intelligence Demand & supply
0.2 Supply Chain X X X X X X
management
Contract
0.3 Automation of Supplier Contract Supply Chain X X X X X X
management
EcoPeer An Industry ESG Supplier Risk Analysis, Supplier Risk
0.4 Supply Chain X X X X X X
benchmarking Analysis
Contract
0.5 Advanced contract analysis tool Supply Chain
management
1 Contact Center - Agent Productivity Uplift

1.1 Conversation Summarization Operations Customer Service X X X X X X X X

1.2 Call Categorization and Classification Operations Customer Service X X X X X X X X
1.3 Sentiment Analysis & Root cause analysis Operations Customer Service X X X X X X X X
Automatic Acknowledgment Email at end of call with
1.4 Operations Customer Service X X X X X X X X
contextual call summary
1.5 Agent Behavior and Call Quality Analysis Operations Customer Service X X X X X X X X
Knowledge Management - Knowledge Search and
2
Knowledge Generation
AI-Powered Similarity Search Engine For Speedier Ticket Knowledge
2.1 Analysis X X X X X X X X
Resolution! Management
Knowledge
2.2 ChatGPT enabled knowledge augmentation from videos Augmentation X X X X X X X X
Management
AI-Powered Auto KB Article generation for Shared Knowledge
2.3 Generation X X X X X X X X
Service/Contact Center Management
Knowledge
2.4 GPT-3 Enabled Earnings Call Transcript Analysis -demo Analysis X X X X X X X X
Management
 Generative AI use cases (from Section GDS DNA Consulting team)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)
3 ESG, Contract Risk management, Risk Management
EcoPeer: ESG Supplier & Competitor Risk Analysis &
3.1 Risk Management Risk analysis X X X X X X X X
benchmarking
Contract
3.2 Advanced Supplier & Legal Contract Analysis Tool Risk Management X X X X X X X X
management
3.3 SmARRF - Smart Auto Risk Register Filling Risk Management Risk Management X X X X X X X X
LLM powered PEP & Background Verification and Due
3.4 Risk Management Verification X X X X X X X X
Diligence (WIP)
3.5 TPRM Risk Management Risk Management Risk Management X X X X X X X X
4 Social Media Analytics and Hyperpersonlization
4.1 Product Catalogue Description Generation Using GPT-3 Sales & Marketing Marketing X X X X X X X X
4.2 Social Media LLM Content Analysis Sales & Marketing Marketing X X X X X X X X
4.3 Hyper personalisation Sales & Marketing Marketing X X X X X X X X
5 Healthcare
Customer
5.1 Claims Denial Rebuttal Operations X
management
Customer
5.2 Claim Denial Email Operations X
management
5.3 MedMind AI Operations Customer Service X

5.4 DoctorAssist Operations Customer Service X

6 HR Generative AI Use cases

Applicant Resume Screening and CV to JD match using
6.1 Human Resources HR management X X X X X X X X
Generative AI
6.2 Voice of Employee Analysis using Generative AI Human Resources HR management X X X X X X X X
LLM powered PEP & Background Verification and Due
6.3 Human Resources HR management X X X X X X X X
Diligence (WIP)
 Generative AI use cases (from Section Europe West Gen AI team)

S.No. Title Demo Video Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT
Supplier negotiation Category Management
1 Procurement X X X X X
cockpit & Supply Chain
Customer negotiation Category Management Customer
2 X X X
cockpit & Supply Chain management
Category Management Supplier
3 Supplier contract analysis X X X X X X
& Supply Chain management
Category Management Customer
4 Customer contract analysis X X X X X X X X X
& Supply Chain management
Assistant for contract
5 All Functions n/a X X X X X
management
Automated supplier inquiry Category Management Supplier
6 X X X X X
handling & Supply Chain management
Automated customer / Category Management Customer
7 X X X
insurer inquiry handling & Supply Chain management
8 Product Data quality IT & Technology Data management X X X X X X
Knowledge
9 Knowledge Database X IT & Technology X X X X X X X X
management
Customer onboarding
10 Sales & Marketing Lead to customer X X X
assistant
Automated ad (-vertising
11 Sales & Marketing Marketing X X
materials) creation
12 Sales content creation Sales & Marketing Marketing X X X

13 Customer service co-pilot Sales & Marketing Lead to customer X X

14 Intelligent Payroll Chatbot X Human Resources HR management X X X X X X X X

15 Sales force assistant X Sales & Marketing Sales strategy X X

 Generative AI use cases (from Section Europe West Gen AI team)

S.No. Title Demo Video Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT
Personalized language
16 Sales & Marketing Sales strategy X X X
stylist
Plan, Budget,
17 AI-tailored dashboards X Finance & Controlling X X X X X X X X
Report (ing)
Quality assurance Operations & Quality
18 X X X X
management Logistics management
Sales strategy
19 Demand forecasting Sales & Marketing X X X X X
& planning
Product content
Dscrp.-X Lead to customer,
20 (description and image) Sales & Marketing X X X
Image- No Marketing
generation
Factory location network
21 Strategy & Innovation Idea to offer X X X
optimization
Category Procurement.
Inventory & procurement
22 Management & Inventory X X X
optimization
Supply Chain management
Large-scale regional Technology & Product
23 X X
success prediction Product development development
Strategy &
24 Internal process efficiency Strategy X X X X X X X
Innovation
Customer landscape Strategy &
25 Idea to offer X X X
analysis Innovation
Lead to customer,
26 Consumer insights Sales & Marketing X X X X
Marketing
27 Urban digital twin Strategy & Innovation n/a x X
Construction multi-
28 Strategy & Innovation Strategy X
dimensional modelling
Automated tender
29 Sales & Marketing Lead to customer X X X X
evaluation
Sales & Sales strategy,
30 Virtual shopping guide X X X
Marketing Lead to customer
 Generative AI use cases (from Section Europe West Gen AI team)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT

AI-based return and accrual Financial report &

31 Finance & Controlling X
prediction audit (ing)

Assortment planning Category Management Inventory

32 X
optimization & Supply Chain management

Category Management
33 In-store-execution Store management X
& Supply Chain
New business models from car
34 Strategy & Innovation Idea to offer X
sensor data
Customized discounts and
35 Sales & Marketing Marketing X X
product promotions
36 Deal sourcing & diligence Strategy & Innovation Strategy X
Sales strategy
37 Drug Demand Forecasting Multiple X
& planning
Research & Product design &
38 Drug Discovery X
development engineering
Drug Development – Research & Product design &
39 X
Reinforcement development engineering
Sales & Marketing,
40 Intelligent dental assistant n/a X
Dental manufacturing
Optimized Net Revenue
41 Sales & Marketing Sales strategy X
Management system
42 Accounting Analytics Finance & Controlling Controlling X X X X X X X X
Transaction Anomaly
43 Finance & Controlling Procure to pay X X X X X X X X
Detection – Procure to pay
Automation of EBIT- Plan, Budget Report
44 Finance & Controlling X X X X X X X X
Forecasting (ing)
Accounts Receivable
45 Finance & Controlling Treasury X X X X X X X X
Collections Assistant
 Generative AI use cases (from Section Europe West Gen AI team)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT

Chatbot for accounting Financial mgmt &

46 Finance & Controlling X X X X X X X X
standards reporting

Financial mgmt &

47 Reports processing Finance & Controlling X X X X X X X X
reporting
Sales Strategy, Lead
48 Translation service Sales & Marketing to customer, X X X X X X X X
Marketing
Software / App
49 GitHub Copilot IT & Technology X X X X X X X X
development
Software / App
50 Code generation IT & Technology X X X X X X X X
development
Category Management &
51 Incoming customer calls Customer Service X
Supply Chain
Category Management &
52 Incoming customer mails Customer Service X
Supply Chain
 Generative AI use cases (from Section Europe West Gen AI team)

S.No. Title Key function Key Sub-Function Segment Sector Function Technologies Source
PPE pre-emptive maintenance -Rail Category Management & PPE pre-emptive Machine
53 B2B Mining Risk Preemption
Network Operator Supply Chain maintenance Learning
PPE pre-emptive maintenance - Category Management & PPE pre-emptive Structural Health Machine
54 B2B Infrastructure
Structural Health Monitoring system Supply Chain maintenance Monitoring Learning
PPE pre-emptive maintenance - Category Management & PPE pre-emptive Machine
55 B2B Robotics PPE maintenance
Production Line Industrial Robots Supply Chain maintenance Learning
Demand Forecasting - Palm Oil Category Management & Supply Chain Machine
56 Demand Forecasting B2B Agriculture
Producer Supply Chain Forecasting Learning
57 Demand Forecasting - Port Logistics Category Management & Supply Chain Machine
Demand Forecasting B2B Logistics
Operator Supply Chain Forecasting Learning
Machine
58 Dynamic Pricing Platform Finance & Controlling Dynamic Pricing B2C Insurance Pricing
Learning
Network usage decision-making – Rail Category Management & Machine
59 Network Planning B2B Mining Network Planning APAC /
track priorization Supply Chain Learning
Oceania
Text digitization by OCR – Income and MS AI Cognitive Gen AI use
Loan application
60 expense verification for Investment n/a Text Digitization Cross Financial Svc EY’s custom cases
processes
Bank NLP modules

61 Text digitization by OCR – Patient Document Machine

n/a Text Digitization B2B Public
questionnaire recording for a hospital management Learning
Reconciliation process for Investment Reconciliation Machine
62 Finance & Controlling Reconciliation Process B2B Financial
Bank process for IB Learning
Machine
63 Customer Credit Risk Management Finance & Controlling Risk Management B2B Financial Risk Management
Learning
Identification of cartographic areas of Cartographic Machine
64 n/a Cartographic Search B2B Public
interest Search Learning
Accounting and Reporting automation Accounting & Machine
65 Finance & Controlling Cost and Capitalization B2B Tech
- Fixed Asset Cost and Capitalization Reporting Learning
 Generative AI use cases (from Section additional slides)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT
Financial &
1 Company sentiment analysis Finance & Sustainability Sustainability X X X X X X X X
reporting
Sales Strategy, Lead
2 Sales agent Sales & Marketing to customer, X X
Marketing
Sales Strategy, Lead
3 Product recommendation Sales & Marketing to customer, X X
Marketing
4 ESG GenAI Benchmarking Sustainability Benchmarking X X X X X X X X

Business model and market

5 Strategy & Innovation Strategy X X X X X X X X
impact assessment
 Generative AI use cases (from Section Oceania and APAC)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT

1 Code conversion Technology Data management X X X X X X X X

2 Test data preparation Technology Data management X X X X X X X X

Business uplift through

3 Technology Automation X X X X X X X X
technology

Responsible AI – Regulatory
4 Technology Regulations X X X X X X X X
Landscape & Risk Assessment

Classification and
5 PoisonGPT Knowledge management X
extraction
 Generative AI use cases (from Section EY Tech Consulting)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT

1 Governance & Risk Mitigation Technology Governance X X X X X X X X

Generative Product Design

2 Product innovation Product design X X X X X X
Assistant
Revenue Growth Management Sales Strategy, Lead
3 & AI - Consumer commercial Sales & Marketing to customer, X X X X X X X X
analytics platform Marketing

Algorithmic Edge Advances

4 Media-Mix Modeling & Scenario Knowledge Management Visualization X X X X X X X X
Planning

Integrating AI across all phases

5 Supply Chain End to end X X X X
of supply-chain
 Generative AI use cases (from Section Insurance)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT

1 Gen AI Driven FNOL

FNOL: Summarization and

1.1 Financial Management Reporting X
extraction: Solution

Gen AI supported FNOL : Data

1.2 Financial Management Reporting X
Flow

2 Gen AI Driven UW

Underwriting : Extraction &

2.1 Knowledge Management using Financial Management Process X
Q&A : Solution
Gen AI Supported
2.2 Financial Management Process X
Underwriting: Data Flow
 Generative AI use cases (from Section SAP EY AI use cases)

S.No. Title Key function Key Sub-Function AM&M Consumer Energy FS G&I (RE) HS&W PE (Fund) TMT
EY Intelligent Transformation
1 Technology Data management X X X X X X X X
Platform
2 EY Accounting AI for Payables Financial Management Process X X X X X X X X

3 EY Intelligent Spend Analytics Supply Chain Procurement X X X X X X X X

EY CFIN SAP Error Reporting

4 Technology Data management X X X X X X X X
Dashboard in SAC
EY SOFIE: Smart Secure
5 Technology Data security X X X X X X X X
Solution
EY Intelligent Warehouse
Warehouse
6 Management – Storage Area Supply Chain X X X X X X X X
Management
Determination
EY SPACES: EY Intelligent
7 Sales & Marketing Customer experience X X X X X X X X
Insights
8 EY Smart Onboarding Human resources Onboarding X X X X X X X X

EY Intelligent Contract Contract

9 Supply Chain X X X X X X X X
Generator management
EY BTC iRPA: Automated A1
10 Human resources Automation X X X X X X X X
and PWD submission
EY Intelligent HelEYos
11 Training existing ML algorithm Technology Data management X X X X X X X X
using genAI
12 EY PUIC – AI DEBT Prediction Financial Management Debt management X

13 EY Power and Utility – Virtual

X X X X X X X X
agent
 Generative AI use cases (from Section EY Generative AI Credentials Pack)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)
1 Generative AI Enabled Product Design Customer experience Access, Analytics X
2 Generative AI Factory and High Value Use Cases Operations Customer Service X
3 Generative AI Center of Excellence and Use Cases Operations Customer Service X
4 Store Ops Assistant Operations Customer Service X
5 Generative AI Chatbot POC Customer experience Customer Service X
Product
6 UX/UI Development for Custom GPT Solution IT & Technology X
development
7 LLM for digitization of rules of the road Customer experience Customer Service X
Product design,
8 Generative AI - Enterprise Strategy & Pilots Technology innovation
X
Operations, Customer
9 Generative AI Factory, Sales Forecasting Tool, and Chatbot Automation X
experience
10 Call Summarization and Insights Capture from Advisor calls IT & Technology Automation X
Global Consumer Bank leveraging Generative AI to Conform
11 Technology, Regulations Innovation X
to Regulatory Feedback
Financial Services HR Department – ChatGPT for Enterprises Finance, Human
12 Automation X
with Guardrails Resources
Generative AI Strategy & Roadmap for a Global Insurance
13
Company Strategy Optimization X
Product
14 OpenAI GPT4 based Solution Financial Advisory Finance X
development
15 AI Governance Framework and Target Operating Model Governance Innovation X

16 Generative AI Operating Model & Roadmap Operations Innovation X

 Generative AI use cases (from Section EY Generative AI Credentials Pack)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)
17 Generative AI and AI Lab Innovation Analytics X
Generative AI Strategy & Governance for a Global Payments
18 Governance, Operations Customer Service X
company
19 Funding Proposal Drafting and Management Contract Management Customer Service X
EY Intelligence Engine Enterprise Asset Management
20 Asset Management Technology X
Solution

21 Generative AI Transformation for MedTech Technology Transformation X

Supplier
22 CSR Generative AI Social Responsibility X
management
Medical Legal Review (MLR) / Promotional Materials
23 Research & development Screening X
Lifecycle (PML)

24 Intercompany Agreements / Contracts analysis Contract Management Customer Service X

25 Commercial Market Landscape Reports IT & Technology Automation X

26 Fortrea Data and AI Strategy IT & Technology Automation X

27 Generative AI Powered Agronomy Assistant Sales & Marketing Customer Service X

28 Generative AI - Conversational Business Intelligence Sales & Marketing Customer Service X

29 Generative AI - Standard Operating Procedure Navigator Operations Optimization X

Knowledge
30 Generative AI Standards Bookshelf Search Operations X
management
Analytics,
31 AI Innovation Studio and AI Innovation Hub Innovation X
Automation
Generative AI Governance, Platform Development, and Use Product
32 Technology X
Case Scaling development
 Generative AI use cases (from Section EY Generative AI Credentials Pack)

G&I PE
S.No Title Key function Key Sub-Function AM&M Consumer FS Energy HS&W TMT
(RE) (Fund)

33 Enhancing Employee and Guest Experience Customer experience Process X

34 Marketing Compliance PoC Sales & Marketing Customer Service X

35 Generative AI Product Knowledge Assistant Sales & Marketing Customer Service X
36 Recursive Story Creation System Knowledge Management Customer Service X
Generative AI Architecture for a Finanical Co-Pilot for a Technology & Product Product
37 Bleeding-Edge Manufacturer X
development development
Gen AI use cases
from Sector GenAI GTM
decks
 Use case category:

Integrating OpenAI into core programs at a G360 Bank

L4: Implemented with client1
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A G360 Bank is executing a program to integrate OpenAI into core platforms with the intent of (i) improving access to internal content by employees
(Financial Advisors, Service Associates), (ii) enhance employee productivity, and (iii) improve client experience.

Value Areas
Use Cases
Internal Knowledge Management ► Lower effort & higher productivity:
Developed Improve access and retrievability of Accessing/Searching for information, tagging
organization’s knowledge base through content, etc.
Internal Knowledge Management Virtual Assistant, enhanced search, and
► Virtual Assistant ► High quality & consistent content: New content
creating quality metadata
► Search generated e.g., knowledge - title, keywords, FAQs,
(Summarization, Taxonomy, FAQs)
► Metadata (Summary, Taxonomy, FAQs) notes, etc.

► Better service & client engagement: Understanding

Call Center Analytics user needs and experiences (sentiment, etc.)
Ongoing
Increase agent productivity and client
Call Center Analytics experience by deriving interaction ► Reusable assets for LLM adoption and scaling:
insights for CRM logging, flagging Testing/ evaluation suite and enhanced LRC
► Call Center to Client governance, etc.
opportunities and risks in conversations,
► Financial Advisor (FA) to client
and better workforce training

EY SUPPORT AREAS

Use Case Design & Model Testing Future State BAU MRM/Legal/Risk Overall Program
Development (Testing Suite) Design Documentation Support

Page 21 Return back to meta data

 Use case category:

LLMs to automate Funding Proposal Drafting in Government

L4: Implemented with client1
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Use Case
EY is helping a state government use LLMs to automate pipeline management and responses to the federal health grant application process by (i) extracting
the key concepts from grant opportunity and (ii) drafting the proposal from relevant content with minimal effort

Matchmaking & Application Human-in-the-Loop Fine- Value Drivers

Tagging
Generation Tuning
► Improved efficiency: streamlines application
Funding Match grants with the process and saves time
Topic 1
Opportunity state
Topic 2
► Consistency and accuracy: Minimizes errors
Topic 3
Display each section, and misunderstandings
smart editor, etc.
Var 1 separately ► Customized suggestions: Generates tailored
Var 2 proposal content
Var 3
User edits each ► Learning from past applications: Informs
Extract Submission Format section through pre- future strategies and best practices
► Table of Contents set prompts &
► Project Summary regeneration ► Capacity building: Enhances ability to secure
Save ► Background funding and address community needs
► Approach Download grant
Variables
► Etc.
proposal draft & submit

Provide novel data-driven

Match keywords from the grant Auto-complete application format
connections between clients with
opportunity with the corresponding using function-specific models &
the highest chance of receiving
health factors based on relevance prompts, increasing output quality
funding and relevant grant
scores and reducing error
programs

Page 22 Return back to meta data

 Use case category:

LLM model for ERP interface in Health Sciences & Wellness

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Organizations require efficient methods for managing their enterprise resource planning (ERP) systems—traditional ERP systems often present challenges to
users such as complex interfaces and the need for advanced technical knowledge to operate them.
ChatERP harnesses the power of Generative AI for intuitive ERP interaction through natural language, enabling seamless data access and updates via chat,
streamlining decision-making and boosting productivity for all users.

Customized Using LLMs as a semantic query layer Custom Answers

U: User C:ChatERP
LLM Model
U Where can I find information on
LLM is instructed Natural language is inventory turnover for product X?
about the used as interface for
C John Doe has written a wiki article
database schema database operations about a dashboard located here
which monitors inventory turnover.
LLM model as Product X turns over 4 times a year
interface for all and has Jane Doe as a product
ERP data Synthesizing ERP with the data lake owner.
U
Update John’s total sales this month
to 300 units.
C
Can you confirm his employee ID
and the product sold?
U
Spreadsheets Emails Dashboard Official Wikis with John Doe, employee ID 1234,
Chat interface containing containing documentation documentation meta-data product X for 300 units this month
built on top of the C
ad-hoc fractional identifying KPIs both internal noting authors I have updated John Doe’s
LLM model analysis documentation and external and history (employee ID 1234) total sales for
Product X to 300 units this month.

Page 23 Return back to meta data

 Use case category:
Option B

ChatERP – enables intuitive ERP interaction, seamless data access and

L4: Implemented with client
L3: Built for EY

updates via chat, streamlining decision-making and boosting productivity L2: Built as a Prototype
L1: Proof of concept

Case example: ERP Management and Understanding: “ChatERP” ILLUSTRATIVE

Problem Statement
Organizations require efficient methods for managing their enterprise resource planning (ERP) systems—traditional ERP systems often present challenges
to users such as complex interfaces and the need for advanced technical knowledge to operate them.
ChatERP harnesses the power of Generative AI for intuitive ERP interaction through natural language, enabling seamless data access and updates via chat,
streamlining decision-making and boosting productivity for all users.

Customized LLM Custom Answers

Model Using LLMs as a semantic query layer U: User C:ChatERP

LLM is instructed Natural language is U Where can I find information on

inventory turnover for product X?
about the used as interface for
database schema database operations C John Doe has written a wiki article
about a dashboard located here
LLM model as which monitors inventory turnover.
interface for all Product X turns over 4 times a year
ERP data and has Jane Doe as a product
Synthesizing ERP with the data lake
owner.
Update John’s total sales this
U month to 300 units.
Can you confirm his employee ID
C and the product sold?
Spreadsheets Emails Dashboard Official Wikis with John Doe, employee ID 1234,
U product X for 300 units this month
Chat interface containing containing documentation documentation meta-data
built on top of the ad-hoc fractional identifying KPIs both internal noting authors C I have updated John Doe’s
LLM model analysis documentation and external and history (employee ID 1234) total sales for
Product X to 300 units this month.

Page 24 20 August 2024

LLMs for Knowledge Management in Banking

Current State Enhanced State

1. Search is limited to a keyword search and creates inefficiencies when 1. LLMs can enhance search by creating richer embeddings for contextual
searching on query context search and accommodating natural language querying

2. Content metadata tends to be lacking and often faces completeness and 2. LLMs can be used for automating metadata generation that can be
accuracy issues reviewed and subsequently published by authors

3. Chatbots built on a set of predetermined FAQs so answers are limited to 3. LLM-powered internal chatbots can be leveraged to analyze content as well
those content areas as find / synthesize the correct answer to user queries

Illustrative Solution Flow:

Better Metadata & Search Synthesized Insights

Question: How do I open a Selected content for

checking account? Q1
Text Generation Q1 Article 2
Model

CV1 Answer: To Open

an account follow
the below steps
Embedding
Model QV1 1. Go to Account
management
.98 Website
QV1 CV2 …

Q1 QV1 CV2
0.1 0.7 0.6 0.8 0.9 0.5
CV3

1 User submits a query Generate & store content metadata; Assess match based on Create prompt with relevant Synthesize response
2 Create embeddings of content (with 3 similarity assessment and 4 content and pass to text 5 from text completion
metadata) - CV & query - QV other relevance algorithms completion

Page 25 Return back to meta data

 Use case category:

LLM for digitization of on-road rules

L4: Implemented with client1
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Use Case
Automated Driving Systems (ADS) must comply with traffic rules and regulations (“rules of the road”). Rules of the road are heavily fragmented, not
harmonized across different jurisdictions, not machine-readable, and open for interpretation.
The LLM for digitization of on-road rules harnesses the power of Generative AI for extracting and formatting all relevant rules of the road within a given
geographic jurisdiction for a given driving scenario.

Collect Scrape traffic rules and Query Ask in natural language Extract Collect relevant sources
regulations from different about the relevant rules of then identify and format
sources and jurisdictions the road to search the requested rules Value Drivers

Output provides  Improved efficiency: Streamlines

expected content rule scouting process and saves
and follows time extracting rules
required structure
 Consistency and accuracy:
Minimizes errors and rule
Query misinterpretations
provides task
instructions  Capacity building: Enhances
ability to scout rules for more
jurisdictions for different driving
scenarios
LLM reads query
and parses text
for answers
Server stores Search ranks most Combined text is
scraped data relevant excerpts used as context

1.As of Nov’23, being implemented with client as a proof of concept

Page 26 20 August 2024 Return back to meta data
 Use case category:

Supplier Contract Analyzer - enables AI-based support for the day-to-day

L4: Implemented with client
L3: Built for EY

work of every CM and automated supplier contract analysis L2: Built as a Prototype
L1: Proof of concept

Case example: Supplier Contract Analyzer ILLUSTRATIVE

Standard components are available “off the shelf” &

don’t need adaptations

Text
Document Information User
to
reader processing interface
text
Business Manage
& &
LLM
prompting retrieve
PDF Dataset logic information XLS
and GPT-4
output model
handling
Input Output

GPT made working with text and language

very easy compared to traditional
complicated and unreliable NLP1 solutions
Application Text
parameter EY-P value-add on proprietary to
handling XLS
components with tailored integration
of business expertise
Standard modules Proprietary modules
1.Natural language processing
Source: EY-Parthenon

Return back to meta data

AI in Multifidelity Drug Design in Health Sciences & Wellness

Problem Statement
Problem Statement: Multifidelity drug discovery often involves a larger dataset of low fidelity screening data and a smaller set of computationally expensive
high fidelity confirmatory data, presenting a challenge in leveraging the available information effectively.
Artificial intelligence can address this challenge by developing models that utilize machine learning techniques to extend the applicability of computationally
expensive Free Energy Perturbation (FEP) methods, enabling accurate predictions and reducing the computational burden in multifidelity drug screening.

Traditional Comp Chem Approach AI Accelerated Approach

Ligand search space Search

(1M ligands) space 2. Transformer based models
approximate FEP function using
1. Low fidelity measurements FEP values and ligand structures
used to pick ligands for FEP 1. FEP is measured
for a subset FEP =
2. Ligands with highest FEP -11.2
recommended for
experimental analysis

3. New FEP values are measured,

The ligand search space is based on Transformer recommendations Active learning with AI
largely neglected by FEP allows a more general probe
measurements of the chemical space

Return back to meta data

 Use case category:

Generative AI-enabled product design in Consumer

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Still conceptual

Use Case
Product design can involve manually intensive design creation. Enabled by AI, however, the entire process can be completed more efficiently, enabling it to
react faster to market trends, improve creativity, and lead to a better design experience.

Benefits Realized
Features
► Concept visualization: Create realistic,
three-dimensional renderings of new
product concepts
25 9K 50K+ ► Design optimization: Analyze the design
data and suggest modifications, creating
better and more cost-effective products
Computer Vision Pre-labeled Non-labeled
Models Images Images ► Material selection: Choose the best
materials based on cost and performance
specifications
Approach
► Prototype acceleration: Render accurate
digital prototypes of proposed product
designs

► Concept evaluation: Perform “what if”

Fine-tune GenAI Work with designers to Implement the solution White space scenarios and enhance the process of
models on data to further refine the model with data integration identification, 3D comparing and assessing product
evaluate the image and make the images and a front-end modeling etc. concepts
generation capability designer-ready interface

Page 29 Return back to meta data

 Use case category:

AI / ML Implementation in Health Sciences & Wellness

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Still conceptual

Use Case
The client wished to set up a center of excellence for implementing automation across various divisions. Too much time was being spent on repeatable,
manual processes that should have been spent on more difficult tasks. The client just had a few RPA pilot use cases prior to the engagement.

Benefits Realized
Initial Assessment
► Improved user experience: Built models
that would improve the health and digital
experiences of customers
1 2 3
► Increased transparency: A more
collaborative and transparent data
No code versioning for the Challenges in obtaining well- No CI/CD style MLOps pipeline science environment
collaborative workspace documented data/features for to productionize
environment for data scientists model generation model ► Enhanced model features: Sourced
features for model build and facilitated
model training and deployment efforts
Approach
using best practices on Azure

► Streamlined AI/ML deployment:

Streamlined the AI/ML development
process by implementing MLOps patterns
Data Science Architecture Feature Store MLOps Chatbot for eventual scaling within the
Used timely insights based on Provided AI, digital, and Made client data Focus on governance, Give members a new organization
client-rich datasets to improve chatbot solution accessible to models for lifecycle management, channel to digitally
digital journeys and health architecture for all efficient training and deployment, and testing of engage with member &
outcomes capability areas deployment in Azure all AI/ML models clinical services

Page 30 Return back to meta data

 Use case category:

RPA Center of Excellence

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Use Case
The client wished to set up a center of excellence for implementing automation across various divisions. Too much time was being spent on repeatable,
manual processes that should have been spent on more difficult tasks. The client just had a few RPA pilot use cases prior to the engagement.

Benefits Realized

 Reduce repeatable processes: The

automations that were developed as part of
the efforts of the center of excellence helped
reduce some of these repeatable processes

 Newtechnologies: The work developed

Intelligent Automation Virtual Helpers for Manual Task Organizational opened the door for new technologies and
Center of Excellence Transactions Reduction Transformation engagements including a large scale chatbot
This led to the The virtual helpers Automations have been Create a positive
implemented have designed to reduce the outlook around  New tools: Started the work at the client with
automation of over 35
processes across completed over 1M number of manual tasks automation with cross- Blue Prism and later introduced additional
multiple functional transactions so far. completed by the divisional change intelligent automation tools
areas including Tax, They have completed client’s team members. management
HR, Finance, Supply over 45K hours of strategies.
Chain and many others. work.

Page 31 20 August 2024 Return back to meta data

 Use case category:

Mixed Media Modelling Optimization

L4: Implemented with client

Media Mix Modeling (MMM) Optimization Approach L3: Built for EY

L2: Built as a Prototype
L1: Still conceptual

A Successful MMM modeling optimization approach leverages generative AI such as advanced scenario modeling, visualization and conversational query for
optimization recommendations. To achieve this it requires strategic planning, stakeholder collaboration, and effective communication. The high-level
activities are laid out below.

EY Enablers
High-level approach
► EY’s MMM
Assessment and
benchmarking
templates
Assess current state Define future state Design high level roadmap MVP/POC
► Example future
► Identify current stakeholders ► Define future state vision of the ► Create suggested list for proof-of- ► Agree on minimum viable product state models from
► Conduct stakeholder interviews to MMM process concept use cases (MVP) or proof-of-concept (POC) prior
understand the current MMM Develop high level Leverage EY use case prioritization scope and design
► ►
engagements
capabilities including business recommendations and technology scorecard model to help prioritize ► MVP/POV model development and
objectives, strategy, and current blueprint focused on MMM use cases testing
tech infrastructure recommendation engine, real-time ► Develop a high-level ► Development of MVP/POC ► Library of
► Develop a high-level understanding campaign optimization capability implementation roadmap to build visualization dashboard for insights example
of current data inputs and outputs and scenario modeling capability the future state focusing on quick Generative AI use
wins and long-term opportunities
cases

Deliverables
► Stakeholder interview summary ► Defined strategy and vision ► List of prioritized use cases ► MVP/POC scope and design
► Current state discovery ► Recommendations based on ► Roadmap for implementing defined requirements
assessment results current state assessments future state ► MVP/POC model and test results
► Defined future state framework ► MVP Dashboard

Page 32 Return back to meta data

 Use case category:

GenAI Asset management solution

L4: Implemented with client1
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

EY Intelligence Engine Enterprise Asset Management Solution

Use Case
Legacy enterprise asset management systems face several challenges, such as the high lead time to get insights, limited scalability and integration, lack of
real-time visibility, and no proactive insights. EY’s Asset Management Solution seeks to solve these problems.

Benefits Realized

Integration of data via data fabric Cross application integration  Document intelligence and actions: Link
Speed up the time to integrate data from Enables integration with other systems like service notes to assets, automatically parse
different legacy systems including Oracle EBS, purchase order management to enable one click documents into structured data
SAP, Maximo order of spare parts
 Interactive map filters: Display assets across
locations on a map for enhanced insights,
filter assets based on geolocation
Alert driven actions Intelligent communications  Integration
with external systems: Allows
Sends persona-based alerts for assets needing Supports human interaction through SMS and ordering parts with a simple click, like a
maintenance including insights on reliability email. Can contextualize worker notes and non- purchase order management system
engineering metrics traditional requests from CEO/CFO/Shopfloor
Manager.  Assetbased benchmarking: Fully customizable
metrics, identify key trends or concerns,
compare vs other assets in class

Chatbot with generative AI Graph based representations

Advanced natural language processing and Infer dynamic relationships, model complex
chats to provide critical asset impact and best asset hierarchies and lifecycles with greater
courses of action flexibility than traditional table-based data platforms

Page 33 20 August 2024 Return back to meta data

 Use case category:

SourceGPT – Bid response assistant for the deregulated markets

L4: Implemented with client1
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

5 Case example: Bid response assistant for the deregulated markets

SourceGPT
ACC E L E R AT E N E G OT I AT E C R E AT E
Time to value Initial terms for sourcing Data for insights
Gen AI for RFP Support and Vendor  Craft a summary of supplier  Manage the initial back-and-forth  Negotiation dataand outcomes
Negotiation proposal submissions to an RFP on behalf of sourcing managers can be leveraged to improve
event. Draft outreach emails and summarize negotiation procurement strategies and
inviting suppliers to negotiations. outcomes for humans to pick up tactical workload management
Demo developed in 10 days to showcase how chatbot Improve RFP event execution and move forward
can support Procurement users to manage the RFP timeframe
process and run vendor negotiations, enabling operations
at scale.

This use case will demonstrate the chatbot’s ability through these steps:
1. Define your procurement need with KPIs
M I T I G AT E OPTIMIZE
2. Upload the RFP responses from different vendors Capacity constraints Communication efforts
3. Review the generated executive summaries for each RFP  Supportfor Procurement
 Reduce effort related to
workloads especially in light of
consolidation of RFP data and
4. Generate and review negotiation strategies supply chain disruptions and key
written communication to
personnel availability risks to
5. Generate, edit, and save the outreach emails suppliers
business continuity

Page 34 Return back to meta data

 Use case category:

EcoPeer– An Industry ESG Supplier Risk Analysis and Benchmarking

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

1 Case example: EcoPeer

Overview
EcoPeer empowers to evaluate ESG performance of an organisation against peers, ESG related recognitions or controversies and key ESG insights from the
company’s annual responsibility report

Business Ask Solution Brief Tools Used

 Evaluate the ESG  Designed and deployed a  GPT3

performance of an Generative AI -based model to  Yahoo Finance
organisation against peers assess ESG performance and
derive insights  ChatGPT
 Evaluate ESG-related
recognitions or controversies  Utilised insights from Yahoo  Hugging face zero-shot
Finance to enhance the classification pipeline
 Drive key ESG insights from
the company’s annual predictions  Power BI
responsibility report  Used PowerBI for interactive
dashboards

Impact/Benefits Realised

Comprehensive view of Strategize to improve Sustainable financing Manage potential Better stakeholder Avoid
ESG performance sustainability decisions made easy reputational risks engagement greenwashing
initiatives

Page 35 Return back to meta data

 Use case category:

GenAI can be leveraged to build product descriptions automatically, optimize

L4: Implemented with client
L3: Built for EY

them for SEO and adapt them specifically to individual customer profiles L2: Built as a Prototype
L1: Proof of concept

GenAI-enabled product description

Cumulative GenAI product description writing use cases

A B C

Product description generation SEO-optimized product listing writer Customer-specific product description
Automated product description to Advanced recommendation based on Persona based, dynamic product presen-
improve information for purchase decision incoming customer feed and SEO tation with pictures and text description

Lorem ipsum dolor sit amet, Etiam vitae tempus lectus.

consectetur adipiscing elit. Donec mattis, felis eu aliquam
Proin sollicitudin lectus risus, tempus, ipsum ligula porttitor
id sagittis massa lobortis sed. nulla, quis gravida leo enim eu
Nulla nec est erat. Sed non mi.
diam ac nisi venenatis
Mauris ex nulla, pretium nec
faucibus. Nullam pharetra nisl
tincidunt eu, rhoncus nec
leo, laoreet dapibus velit diam
purus.
aliquet odio.

Based on customer specific characteristic,

e.g. age, level of education etc. product
descriptions will be personalized, e.g. short
vs. long text, language, writing style etc.

D Monetization opportunity: Offer GenAI-enabled product description as a service to platform partners

Source: EY-Parthenon project expertise

Page 36 Return back to meta data

 Use case category:

Generative AI allows to create and customize product descriptions at scale, L4: Implemented with client
L3: Built for EY

establishing best-in-class product information transparency for customers L2: Built as a Prototype
L1: Proof of concept

Product description generation USE CASE | ILLUSTRATIVE

Product description generator Deep dive – Information content generation

Company

Poor product information transparency leads to higher

Problem
risk of product returns and low conversion rate c Company

Product master data Existing images / visuals

Generative AI to automate the creation of product
Solution descriptions at scale with content customization in
different formats (e.g. text, images, etc.) or languages
(Pre-trained) AI model uses real
images and metadata to generate
Realized  Enhanced customer shopping experience and content in the desired style
business customer satisfaction
impact  Increased conversion rates

Generative AI text model Auto-generated descriptive

creating info / marketing product images based on
Offensive
Enabling new
business model

Impacted Sales & Marketing Business model

integration
Scaling of
content for online shop product information
business IT & Technology capabilities
functions Defensive
Protecting existing
business model Cost optimization

Value levers
Revenue growth

Best-in-class product
Solutions information transparency

Source: EY-Parthenon project expertise

Page 37 Return back to meta data

 Use case category:

Already today, LLMs are time savers being capable to create high-quality
L4: Implemented with client
L3: Built for EY

article descriptions for your web shop via data-to-text generation L2: Built as a Prototype
L1: Proof of concept

Product description generation DEEP DIVE | ILLUSTRATIVE

“Write a brief product

Sizing Information Runs small, please order one size larger description based on some of
GPT 3.5 the following descriptions...”
Color Bordeaux

Look Contrast-colored details AI-generated description Real description

Upper Material Leather, Textile Discover the remarkable Whenever you wear it, you're
Bordeaux Contrast Sneakers, sure to get compliments: the
Style Sporty where style seamlessly meets "ML373" sneaker from New
functionality. Meticulously crafted Balance is sure to impress with its
Closure Lacing
from a blend of leather and shimmering metallic details right
Closure Details Adjustable textile, these sporty shoes from the first glance. In addition,
showcase adjustable laces for a it features a padded entry and
Toe Shape Round
secure fit, while the advanced soft leather upper. The EVA
Cushioning Technology EVA EVA cushioning guarantees midsole of the casual leisure shoe
exceptional comfort. Enhanced by ensures perfect cushioning, while
Outsole Material Rubber
modern design elements, these the slip-resistant rubber outsole
Outsole Properties Slip-resistant sneakers perfectly complement stands for optimal traction.
athletic pursuits and urban Always the first choice with a
Outsole Profile Lightly profiled
exploration. shirt and jeans - the stylish
"ML373" sneaker from New
Balance.
Source: EY-Parthenon

Page 38
 Use case category: Option B
Already today, LLMs are time savers being capable to create high quality article
L4: Implemented with client
L3: Built for EY

descriptions for your web shop via data-to-text generation L2: Built as a Prototype
L1: Proof of concept

X Case example: Product description generation – Example DEEP DIVE | ILLUSTRATIVE

Product ID 9492810

Product name Weber Gasgrill Spirit E-330 Classic

AI generated description
Real description
The Weber Gasgrill Spirit E-330 Classic
The Weber gas grill Spirit Classic E-330
Product category Gas grills is a must-have for any grilling
GBS is a comfortable grill cart that has
enthusiast. This grill wagon features
many extras ready. Venture with the
three burners and a side burner,
gas grill to new experiments and
 Enameled lid with lid providing ample cooking space and
expand your grilling skills. With extras
thermometer: for optimal versatility. The sear zone allows for
such as the Sear Zone, side stove and
temperature control high-temperature grilling of steaks,
GBS grill grate, you now have even
Product details while the modular Weber GBS-grillrost
 Fixed side tables with 3 cutlery more room for enjoyment and
system enables a variety of cooking
hooks each experience unlimited grilling options.
options. The integrated thermometer
… Thanks to the Weber flavor rails and
ensures optimal temperature control,
ideal heat distribution, 100% delicious
and the crossover-piezo ignition system
grilled taste is guaranteed every time.
ensures easy and safe ignition.
 Sear zone for perfect steaks …
…
Marketing  GBS grill grate for multiple
message grilling methods
…

Source: Desk research; EY-Parthenon project expertise

Page 39
 Use case category:

EY’s Generative AI Product Listing Generator automatically creates new

L4: Implemented with client
L3: Built for EY

content across large portfolios in a fraction of time of the manual process L2: Built as a Prototype
L1: Proof of concept

Case study: Generative AI Product Listing Writer

 Represents the voice of the customer: Builds product descriptions that represent the voice of the customer in a dynamic
Value way
drivers  Reduced time for content creation: Creates content across large portfolios in a fraction of the time of the manual process
 Improved keyword generation: Introduces keywords in the product reviews that help with search engine optimization

Source: EY-Parthenon

Page 40 Return back to meta data

 Use case category:

Generative AI Product Listing Writer - uses basic product information to write

L4: Implemented with client
L3: Built for EY

compelling product listing pages in the customer’s voice L2: Built as a Prototype
L1: Still conceptual

X Case example: Generative AI Product Listing Writer

Problem Statement
EY’s Generative AI Product Listing Generator can automatically create new content across large portfolios in a fraction of the time of the manual process.
The tool uses product reviews to build descriptions that represent the voice of the customer in a dynamic way.

SEO Comparison of Water Purifier Descriptions

Value Drivers
GPT Written + Reviews
Current GPT Written
Retraining  Represents the voice of the customer:
Builds product descriptions that
14 relevant keywords 25 relevant keywords
5 relevant keywords represents the voice of the customer in a
(3x) (5x)
dynamic way
P&G Purifier of Water is a portable water Introducing the P&G Purifier of Water The P&G Purifier of Water Packets is the ultimate
treatment system that you can take Packets - the ultimate survival water solution for safe drinking water in any outdoor
anywhere. Whether you are enjoying the treatment solution for camping and adventure. You can be assured of clear water any  Reduced time for content creation:
outdoors, traveling to remote countries outdoor adventures. With 240 packets in time, anywhere. This camping chemical water Creates content across large portfolios in
or working in disaster relief, you will each pack. you can easily purify large treater is an essential for emergency preparedness
never have to worry about bringing volumes of water, making it safe to drink. and ultimate survival water treatment. Our water
a fraction of the time of the manual
enough water. These small and Each packet has 99.99% of bacteria purification packets remove 99.99% of bacteria, process
lightweight travel packets are the removed and provides you with clean and ensuring that the water you drink is clean, safe and
ultimate survival water treatment that safe water to drink. Our mission is to healthy. It also enhances the taste, making it
can treat 10 liters of water in just 30 provide clean water sustainably in the enjoyable to drink. Our mission is to provide clean
 Improved keyword generation: Introduces
minutes. global south. Their effectiveness has water sustainably. You can enjoy the great keywords in the product reviews that
been proven through rigorous testing. outdoors without worrying about the quality of the helps with search engine optimization
These packets are perfect for those who water you drink. This camping chemical water
love hiking. Don't let contaminated water treater is a must-have for anyone who loves the
ruin your outdoor adventure. outdoors. It's easy to use, effective and reliable.
You have access to safe drinking water and get to
experience the peace of mind that comes with
having the best water purification system on the
market.

Page 41
 Use case category:

AI-driven algorithms empower companies to make data-backed decisions by

L4: Implemented with client
L3: Built for EY

providing early, actionable insights into consumer demand L2: Built as a Prototype
L1: Proof of concept

Trend based curation via demand sensing USE CASE | ILLUSTRATIVE

Consumer insights Deep dive – understanding consumer behavior

Consumer insights Consumer profiles

Consumer trends, preferences and segmentation
Problem are highly dynamic and require large amount of Global web Product reviews
crawling of (social) and related
resources media comments

Use AI capabilities to collect consumer data, Buyer information Buyer behavior Transactional data
(demographics, etc.) (search activity) (purchases, etc.)
analyze consumer behavior, identify buying
Solution
preferences, and forecast global shopper trends Data input through
based on real-time global data retailer information

Global data is processed through global

 Earlyidentification of relevant trends Historically collected buyer information
language competencies of AI, analysis of
Realized  More effective marketing and promotion social buzz conducted, and respective
and transactional data is processed

business campaigns trends in countries / regions and related

consumer preferences are identified
impact  Improved and consumer suiting product
presentation Creation of consumer profiles / personas
 Identification of consumer segments
Based on consumer preferences and (e.g., demographics etc.)
evaluated reviews, preferred products and  Recognition of suitable products for each segment
Offensive

Impacted Sales &

Enabling new
business model

Self- brands are identified  Filled out standardized consumer templates

business Key Business model
integration developed
functions Account growth
Globally and locally optimized Identification of trends suitable for each
Defensive
Protecting existing
business model Cost optimization Revenue growth

Value levers

product presentations consumer (segment) and optimized consumer

Solutions targeting
Source: EY-Parthenon

Page 42 Return back to meta data

 Use case category:

A full end-to-end use case is created as multiple applications are combined –

L4: Implemented with client
L3: Built for EY

e.g., complete generation of new landing pages to address recent trends L2: Built as a Prototype
L1: Proof of concept

Auto-generated landing page

Generate Generate Assortment SEO-optimized content Final landing page /

Identify trends
textual content visual content selection and product attribution approval

100% GenAI-generated

Combination of several use cases, namely …

… Demand sensing
… Product description generation
… Picture generation
… Assortment selection
… SEO optimization
… Coding and styling (based on existing style guides)
… will unlock new platform-potential

Page 43
Source: EY-Parthenon Return back to meta data
 Use case category:

GenAI can provide context-driven insights e.g., analyze supplier contracts

L4: Implemented with client
L3: Built for EY

and customer demand to support procurement with fact-based negotiations L2: Built as a Prototype
L1: Proof of concept

Use case combination – Fact-based negotiation

Supplier contract analysis Customer lens on supplier proposition

 Provide relevant contract details in a timely fashion  Determine market sentiments: Negative
sentiments for a product can be a leverage point
 Identify unfavorable clauses (e.g. that are
for better terms
inconsistent with standard terms) assisting in
negotiations for better terms  Gain
insights into product features: Negotiate for
products that better align with market demands
 Benchmark contract details against industry stan-
dards or historical data to ensure competitive terms  Benchmark supplier with competitors based on
product reviews …
 Data-driven insights
 Supplier profiling
 Context-driven
exception reporting
 Predictive analytics /
forecasting trends
 Scenario modeling
(prepare for different
outcomes)
Fact-based negotiation support  Risk assessment
GenAI can analyze vast amounts of (contract) data to provide actionable insights and trends, enabling
the procurement team to negotiate with suppliers based on concrete facts and historical performance
Source: EY-Parthenon

Page 44 Return back to meta data

 Use case category:

EY Modern Finance– Enhancing the way financial metrics are interpretated

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

2 Case example: EY Modern Finance

MODERN FINANCE
EY successfully integrated Azure OpenAI in Modern
Finance solution incorporating Sensitivity Analysis and
Financial Insights (“Corporate Analytics”) and
integrating OpenAI GPT-4 features to Enable business
users to ask questions in natural language

Objectives of the OpenAI GPT-4 integration into MF

Suite of models that provide To incorporate OpenAI Deploy this new

more insights related to GPT-4 features for capabilities into the
sensitivity analysis & more complex and current Modern Finance
financial interpretation insightful analysis Solution

Page 45 Return back to meta data

 Use case category:

Wind Technician AI Assistant

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

3 Case example: Wind Technician AI Assistant

Overview
This is a handy tool for field technician to enhance efficiency by providing answers related to standards and procedures on the go, drastically reducing the time
required for searching while ensuring stricter adherence to SoPs

Why do I need to produce work reports

manually?
Why can I not have all the technical
information I need presented to me on
a hands-free device?

Page 46 Return back to meta data

 Use case category:

Sentiment search – Powered by Azure GenAI capabilities to improve the

L4: Implemented with client
L3: Built for EY

employee experience by reducing time spent sifting through documentation L2: Built as a Prototype
L1: Proof of concept

4 Case example: Standard bookshelf

Problem Statement
 A large American utility relied on a legacy knowledge management (KM) system to store standards documentation. This system lacked both consistent indexing and an
effective search capability. Clients initial attempts to roll out new KM capabilities failed due to complexity of the content and the common search problems of
synonyms and polysemy (words with multiple meanings).
 To remain cost-effective, EY had to develop a solution that could search the documents without relying on indexes.

EY Approach Text scraped from

documents and
 EY designed a solution that leveraged the normalized

6,000
advances in Generative AI and an existing virtual
agent to provide an Azure-hosted semantic
search experience. Text embeddings
 The solution automatically created embeddings* created
from thousands of documents. documents searched by
 Users would search for similar documents by meaning improving employee
entering their plain-text prompt into the chatbot Text embeddings stored experience and leading to
on their computer or mobile device. Text embeddings retrieved better, more informed
 The virtual agent would forward the request to an decisions.
Azure OpenAI service and return the most similar
results. Search embedding Similar texts
 Semantic search solved a key business problem created identified
while proving the value of OpenAI, leaving behind
a rich backlog of generative AI use cases.
User enters search
prompt

Page 47 Return back to meta data

 Use case category:

AI Factory Concept: Single platform to deploy AI at scale

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Business Engineering & Enablers Single Unified Architecture

Customer Human + HR
Revenue Operational
Service & Automated
Back Office
Employee
ROI Prioritization
Growth Efficiency (Legal, Risk)
Ops Code Gen Productivity
AI-first Design
Rapid deployments
EBI / Engineering Synergy
Hybrid Agile AI Infrastructure
Unified AI oriented architecture

Secure Enterprise Controls

 AI-Centric Governance / Policy  AI Scale Planning/Management
 Centralized AI Risk Controls  Hybrid Infrastructure Models Talent Pools & Personas
Agile AI
 Automated Code Development Deploying AI  Future Proofing Loosely Coupled & Flexible
Applications
 Continuous ML/Code Pipelines  Extensibility
Human-in-the-loop Integrations
 Data-Asset Controls  AI-Compatible Integrations
AI/ML Specific Integrations
Adaptive Intelligence Layer
Decisioning &
RAG / Fine-Tuned / Custom Trained Language Model(s) Custom Data Models
KPI Metrics / Feedback-Loops

Data Governance
Hybrid Data Sources
Internal Data | Partner Data | Third-Party Data Scalability Planning

Client Zero Ecosystem Model Legal & Compliance

Alliance Partners [i.e. Dell, NVIDIA, Microsoft] + Future Opportunities

Source: Dell-EY Alliance

Page 48 Return back to meta data

 Use case category:

Staffing Model - Pod Structure

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Mature Data and AI Innovation Lab Pod  Each AI product often requires a delivery team
with dedicated resources. Some resources can
Product Manager
serve as part-time to support multiple AI
products

 We see Integrated purpose-built AI agents as

Intelligent Agents – (AI) Product Owner
part of our staffing model

 Business and Digital should maintain alignment

Prompt Subject Matter through a shared product vision and cross-
Engineer Resources functional team composition
AI
Innovation  Function-specificexpertise is critical to
Lab Pod successful AI product delivery
Data Engineers Functional
/ Architects Leaders  Standardized communication and processes
support incremental value delivery
BI Engineers /  The organization model is designed to be
UX/UI Designer
Data Scientists sustainable and industrialize AI products at scale

 Allagile product teams must conform to the data

Quality Assurance (QA) and analytic strategy and standards to drive
Legend:
agility in the AI products and insights
Digital Resource Business Resource

Source: Dell-EY Alliance

Page 49 Return back to meta data

 Use case category:

GPT-4 powered chatbot for payroll inquiries

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

EY developed a chatbot-based proof-of-concept (PoC), leveraging Azure OpenAI’s GPT APIs, for EY
Use Case and Microsoft’s Global Payroll Operate (Tax) to streamline the process for employees to receive
responses to their payroll-related inquiries.
Benefits

Prepare chatbot Utilize chatbot Improve chatbot  Accelerated response

times: Reduction in
Ingest: Employees can inquire the chatbot Human-in-the-loop incorporated response delays due to
on any topics related to data through a metric dashboard to drive limited resources
1. Country-specific ingested by the bot actionable improvements to the
payroll topics chatbot  Ability to address all
2. General payslip questions: Reduction in lost
structure Chatbot experience currently inquiries as all are
entails: immediately answered and
3. Company
specific payroll tracked through chatbot
Multi-Language Capability
policies/FAQs history
4. Connect to  Provides valuable insights:
relevant User ability to give feedback in One source of truth for
systems UI
employee inquiries and
 Chathistory of user feedback by answers for bot to showcase
source documentation important trends and
Train chatbot: Increased trust through
Confidence Score and Source  Summary of user ratings and written patterns
1. Employee Document Viewer feedback
inquiries  Optimize employee
 Resolution rates
2. Prompt resources: Human
engineering Multiple answer display  Response times resources are now able to
options  User volume metrics utilize time for more
strategic work
Source: Dell-EY Alliance

Page 50 Return back to meta data

Customer Service – Blending virtual and human assistance in customers support
experiences

01 02 03 04

Customers Contact Contact Management Augmented Humans

Residential/B2C Customer contact through various GenAI powered virtual agents that Handle inquires that were not
channels such as voice, text, e- can respond to customer requests addressed or partially addressed
Enterprises/B2B
mail, web... and totally/partially replace by virtual agents. Agents use
agents. AI/GenAI tools to process
information, discover context and
make recommendations

Page 51 20 August 2024 Return back to meta data

Gen AI use cases from
GDS DNA Consulting
ChatGPT enabled banker knowledge augmentation from videos

Root Cause
Call Call Sentiment Behavior/Quality Automatic Knowledge
Summarization Analysis of KB generation
Categories Classification Analysis Analysis Agent Response augmentation
Sentiments

Problem Statement
 Contextual Knowledge generation, summarization & extraction for
specific activity within a long video is typically a time-consuming
task. It requires the person to playback the whole video and toggle
to get to the exact point in time to extract contextual information
required while troubleshooting core banking resolution steps

Proposed Solution
 UtilizingChatGPT & OpenAI , we can do semantic search on search
passage & extract relevant snippets( smaller video sections) & key
frames/screenshots and generative summary along with steps to
assist advisor trouble shoot the given problem

Benefits
 Thiscan help save lot of manual work by Shared services /contact
center teams Screen shot Layout

Page 53 Return back to meta data

Asset name: Supplier and Demand Intelligence - UK mineral and mining client (pilot)

A. Information Foundation B. Develop Correlations C. Predict Outcomes D. Enhance Experience

 Establishknowledge graphs for  Scenario analysis based on  Enhanced experience with natural
focus commodities – Scoring of the materials with language-based query handling
– Trade flows suppliers, quality of material cost
– Factors affecting demand and trends, alternative options,
supply patterns demand positions, external
– Impact of macro economic trends, contracts …
indicators
 Trade offs on
 Track suppliers, demand & supply of – Impact of geopolitical events
– Time to buy
critical chemicals to reduce risk – Global and local inventory build
– Rate to buy
 Use this to predict price, up patterns
– Time indexed volume
availability, demand, supply ahead – Supplier dynamics and ranking
by weeks – Risk analysis with internal  Contingency planning
demand trends

Page 54 Return back to meta data

Supply Chain Contract solution to determine payment term risk in contracts

Sector:- CPR
Use case Name:- Contract-
FOP :- Supply Chain – Contract Management
Automation of Supplier Contract Solution Brief
Payment Terms Risk –
Designed & Deployed Contract Assessment Solution
Pilot large CPR UK Client
AI based solution addresses the need to identify
and understand Payment terms risk in the Supplier
Supply Chain Contract solution to contracts
EY helped in implementing a well-designed protocol of
determine payment term risk in
AI/ML driven Supply Chain Platform and System.
contracts

Region – APAC
Business Ask How did we contribute Impact/ Benefits Realised
Technology footprint
Business wanted the solution to -  End to end implementation for  Accelerate Supplier Contract Analysis RPA, NLP, AI
 Performed locally in multiple markets extracting contracting terms from the while managing agreement risk
contract documents and comparing the  Determine best agreement terms and Service Offering
 Reduce manual effort in identification of
payment terms with standard contracts conditions to achieve enterprise goals & Data & Analytics
relevant vendor and contracts and highlighting the associated risk realize benefits
 Automate manual review of complex GDS Penetration :- 90%
 Implemented the use case using RPA,
templates and the contracts terms Artificial intelligence (AI) and Natural Stakeholders :–
Language Processing (NLP) in order to
understand Contract Clauses in context
to buyer/supplier and other Key Contacts
relationships

Page 55 Return back to meta data

Asset name: EcoPeer An Industry ESG Supplier Risk Analysis, benchmarking

EcoPeer empowers to evaluate ESG performance of an organisation against peers, ESG related recognitions or controversies and key ESG
insights from the company’s annual responsibility report
Tool Features
Comprehensive view of ESG performance

Strategize to improve sustainability

initiatives

Sustainable financing decision made easy

Manage potential reputational risks

Better stakeholder engagement

ESG Rating with ESG Initiatives ESG Benchmarking across peers

Avoid Green Washing

Tools & Technology Used

 GPT3  Huggingface  Power BI
 Yahoo Finance zeroshot
classification
 ChatGPT
pipeline

Page 56 Return back to meta data

Asset name: ChatGPT based Advanced contract analysis tool

Uploading contract Pre-processing the Find entities in the Business Benefits

Advanced analytics User Q&A  Automated reading and analysis of
PDF data contract
complex contract documents
Using NLP and GPT3  Better UI/UX enabling great user
Power app triggers The metadata API find clauses in experience
User logs in to the Answer complex Automated notification to users
the azure function to information the contract, 
power app and user queries Informed decision making for
read the contract including entities is financial leakages 
uploads the contract pertaining to business users, by identifying –
and store the data in recognized in the and compare
to be analysed contract using AI  Entities
digital format contract contracts for SLA
breach  Clauses
 Financial Leakages
 SLA Breach
 Pre-validation, negotiation and post
Ingest Extract Identifying Analyse contract Answer validation for the contracts for
Data Data entities, based on different complex user detailed analysis of contracts during
clauses phases questions different phases of contract creation
Read Form Identifying Using GPT3 to Answer
document recognizer entities and analyse contract in complex
clauses in the pre-stage, contract
Ingest data OCR Contract negotiation and queries for Advanced
post validation SMEs contract
phases for analytics
financial leakages,
Power App Blob GPT3 API Azure ML SLAs etc.,
Azure Cloud

Page 57 Return back to meta data

GDS LLM demo catalogue /some are pilot in progress (WIP)

Domain Use Cases

Call Categorization & Sentiment Analysis &
Conversation Summarization
Classification Root cause analysis
Contact Center - Agent Productivity Uplift
Automatic Acknowledgment with
Agent Behavior & Quality Analysis
contextual call summary

Cognitive Search on transcripts Semantic Search on Documents

Knowledge Management – Knowledge Earnings Call Transcript
Search & Generate Knowledge Analysis & Summarization
Create Knowledge Base Articles Document Search on video content

EcoPeer - ESG Supplier Risk Supplier Contact analysis, comparison

Analysis & benchmarking & Contract Risk
ESG, Contract Risk Management, Risk PEP screening, Background
Management verification & Due diligence
SmARF (Smart Auto-Risk Filler) TPRM Risk Management

Social Media Analytics & Hyper- Social Media Product Hyper-personalisation Product Catalogue
personalization Content Analysis offers based on Customer persona and behaviour traits Description Generation

MedMind AI - Assistance for medications and DoctorAssistAssistance for

Healthcare Generative AI Use cases Claims denial rebuttal
lab test information Disease diagnostics

HR Use Cases CV Search (WIP) CV to JD Match (WIP)

Page 58
Contact Center - Agent
Productivity Uplift

Return back to meta data

Use Case – Conversation Summarization

Problem Statement
 Contact service centre Agent has to do post call work, which
involves summarization of call in case management tool.
 It is time consuming and involves at least 2-4 mins of Agent time
and some key information can to be missed by agent while
summarizing the issue

The Solution
 Uses
latest Large Language models like Open AI GPT 3 to create
automated Summary of call transcript

Benefits Screen shot Layout

 AHT Reduction, Error rate reduction, Productivity improvement.

Page 60 Return back to meta data

Use Case– Call Categorization and Classification

Problem Statement
 Person working on client assigned work/call, must correctly identify
whether it is complaint or Request or query. Sometimes it is quite
complex, and there could be penalty applicable for incorrect
classification or handling of customer complaints. There could be
Ombudsmen penalties. It also adds to Cognitive stress of agents

The Solution
 We leveraged Open AI /Large language models to accurately
categorize customer call.
 It assign if the call is complaints, request or the issue
 It further helps do granular level classification into issue
category/subcategory

Benefits Screen shot Layout

 Saves time of Agent, reduces Cognitive stress, Reduces penalty
Risk,

Page 61 Return back to meta data

Use Case– Sentiment Analysis & Root cause analysis

Problem Statement
 Many time contact center representative are not able to interpret
client sentiment and what part /aspect of conversation caused
sentiment to negative or positive.

The Solution
 Solution-Sentiment analysis evaluates the user’s sentiment during
conversation. This enable agent to effective customer
management.
 Root cause analysis of the conversation will drill down on the
reason for sentiment and helps call center to improve their service
to help Coach them.

Benefits Screen shot

 Provides Sentiment and tone, it helps Agent understand the same
and show empathy with client to help improve customer
satisfaction/NPS.

Page 62 Return back to meta data

Automatic Acknowledgment Email at end of call with contextual call summary

Problem Statement
 In
contact center at the end of call, person working on case has to
generate manual acknowledgement email to client covering call
context, steps taken and next steps. This is manual work and
requires Agent time post call ( 1-2 mins of Agent’s time)

The Solution
 leveragesstate of the art ChatGPT/OpenAI to generate automated
personalized acknowledgement email just like humans summazing
issue and next steps

Benefits Screen shot Layout

 Fast
wrap of call, reduction in manual post call work , Personalized
messages help build better customer experience.

Page 63 Return back to meta data

Use Case– Agent Behavior Analysis, Call Quality Analysis (QA) &
Compliance checks

Problem Statement
 QA Department in most of call center have to manually review 4-
5% sample calls (20-30 calls per Agent per month) and have to
listen to calls to do Agent QA and compliance Assessment. Manual
job is time and effort intensive and is error prone.

The Solution
 We leveraged Open AI GPT-3 model to analyze call transcript
between Agent and Customer to evaluate and analyze Agent traits
and call handling against company set guidelines.
 Behavior/Quality analysis automatically evaluates Agent call
Quality on multiple aspects, plus check if conversation was in
compliance with Controls in place.

Benefits
 This allows shared Services to coach agents for better customer
service
 Automation of QA Function.
 QA can be done on 100% sample size rather than 5% sample, this
ensure better compliance

Page 64 Return back to meta data

Knowledge
Management and
Generative AI
Use cases

Return back to meta data

AI-Powered Similarity Search Engine For Speedier Ticket Resolution!

Problem Statement
 Similar or the same issues are raised or logged as support tickets
repeatedly. Worldwide research has found that more than 13% of
all incidents are repeat incidents. Just like with repeat incidents,
organizations are struggling with problems being logged multiple
times and handled in parallel. More than 17% of the total ticket
volume are duplicates. (on top of the 13% of repeats which are
tickets reporting the same problem
 Missing Knowledge customer service representatives have to
often depend on documentation or other seniors for knowledge to
ensure that ultimately, they provide high quality of resolutions!

Proposed Solution
 Provides semantic search engine that automatically identifies
tickets/cases with relevant contents from historical databased of
case records that are contextually similar to query asked. It helps
Agent get access to steps some other agent followed for solving
that type of ticket, helping him get access on stps

Screen shot Layout

Benefits
 Higher FCR, Lower AHT, Less Repeat Issues

Page 66 Return back to meta data

ChatGPT enabled knowledge augmentation from videos

Problem Statement
 Contextual Knowledge generation, summarization & extraction for
specific activity within a long video is typically a time-consuming
task. It requires the person to playback the whole video and toggle
to get to the exact point in time to extract contextual information
required while troubleshooting core banking resolution steps

The Solution
 UtilizingChatGPT & OpenAI , we can do semantic search on search
passage & extract relevant snippets( smaller video sections) & key
frames/screenshots and generative summary along with steps to
assist advisor trouble shoot the given problem

Benefits
 Thiscan help save lot of manual work by Shared services /contact Screen shot Layout
center teams.

Page 67 Return back to meta data

Document Search with Azure OpenAI enables users to search by meaning, innovating
beyond traditional keyword search

Instead of poring through documents looking for key information by title or text, embeddings1 enable users to search by meaning.

Search embedding User enters search

created prompt

Text embeddings Text embeddings

stored retrieved
Text scraped Documents Similar texts
from documents Text embeddings
normalized created identified Results returned
to user

Key Security Considerations Pricing

 Prompts (like those used to generate embeddings) are logged to monitor for  Pricing is incurred when an embedding is generated and is incurred in three
abusive use and to develop and improve the quality of Azure OpenAI’s main areas: (1) initial training, (2) re-training, (3) user search.
content management systems
 It costs $10 for every ~392,000 words that are trained. To process
 Certain customers with low-risk use cases may opt-out of Microsoft's 100,000 documents 1,000 words in length would incur a one-time cost of
retention of prompts ~$2,550.
 Files submitted for fine tuning of models are not retained or used by  Assuming 50 users, if each users submitted 100, 10-word searches
Microsoft to train or improve any Microsoft models monthly, the cost of search would come to $15.30/year.

1.Embeddings are numerical representations (vectors) of the meaning of text

Page 68
High Level Diagram – Enterprise Knowledge Search Q&A &
Knowledge generation

Page 69
CAI Architecture

User interface Language processing & response generation Back end

INTERACTION CHANNELS NLU ENGINE DIALOG MANAGEMENT

Intent Dialog
Entity Extractor Context
Classifier Management
REST
Core HRIS
User Feedback Slots ODATA

SOAP Others

Web apps/ Portal

Transactional
AUTHENTICATION
Core Orchestration Engine use cases
CHANNEL MANAGER
(Custom/ VA adapter - .NET, Node.js, Python)

Live Agent Knowledge Repository

COGNITIVE SERVICES ANALYTICS

Long Tail
OTHERS
Cognitive Search Contextualisation Bot Analytics

Text Analytics Personality Propose Improvements

SQL
Personalization Speech to Text Quality Assurance

Text to Speech

Exception Handling/ Escalation

Bot fallback Search Clarifying dialogue Self-service routing

** Note - This architecture needs further refinements, if it is a bot to bot use case

Page 70
CAI Architecture – Open AI enhanceents

User interface Language processing & response generation Back end

INTERACTION CHANNELS NLU ENGINE DIALOG MANAGEMENT
Dialogue Creation
Training
Intent Entity Dialog
Data Context
Classifier Extractor Management
Generation Multi-Lingual Chatbots
REST
Core HRIS
General Market
Summarization User Feedback Slots ODATA
Intelligence Research Multi-Modal Inputs

SOAP Others

Web apps/ Portal

Transactional
AUTHENTICATION
Core Orchestration Engine use cases
CHANNEL MANAGER
(Custom/ VA adapter - .NET, Node.js, Python)

Live Agent Knowledge Repository

COGNITIVE SERVICES ANALYTICS

Long Tail
OTHERS
Cognitive Search Contextualisation Bot Analytics Next Gen Knowledge
Live Coach, Next best Search
actions
Text Analytics Personality Propose Improvements
SQL
Personalization Speech to Text Quality Assurance
Agent Assist Solution Generate
Text to Speech Content/Analysis
Intelligent Insights

Exception Handling/ Escalation

Bot fallback Search Clarifying dialogue Self-service routing

** Note - This architecture needs further refinements, if it is a bot to bot use case

Page 71
Knowledge Search & Generative AI Q&A Architecture Pattern

Azure Cognitive Search

Cognitive
Open AI Search
Redis Pinecone
Normalization and Cranking layer Embeddings generation Vector generation
Vector Similarity
Un-structured Embedding 1
Data Indexer
DB and Indexer Matching
Search
Embedding 2
Vector Results
Structured
Data Indexer
Documents Documents Embedding 3 match
normalized Cranking
Passage chunking Vectors

Convert search query to embedding Input chunks for

Question Generative AI
Query Embedding Question
API
SQL Connector Prompt to get
Generate Combine relevant relevant text

Answer
Question

Output text across Indexed

passages
Prompt Prompt Map Reduce
Query or question Prompt Engineering
Langchain Pipeline / GPT-Index
Input
Generative AI Response Generative AI Response ( Summarization, Analysis etc)
Business Final
User Search Results, Source Passages, page numbers
Application Output
Middleware Answer

Page 72
Pipeline Combinations we can leverage

Option 1 Option 2 A
MS Azure Search Generative AI Open AI Embeddings + Azure Generative AI
Function+ Azure Form Open AI prompt Engineering,
Recognizer+ Redis Search + python
Input Indexing, Cranking, Search Langchain, Custom Pipelines Similarity Search

Pipeline – Custom python, Azure Langchain, Custom Python

data factory for Connectors, Pipelines
triggers, & Indexing

Option 2B Option 2C
Open AI Embeddings + Azure Generative AI Open AI Embeddings + Azure Generative AI
Function+ Azure Form Open AI prompt Engineering, Function+ Azure Form Open AI prompt Engineering,
Recognizer+ Pipeline + Similarity python Recognizer+ Redis Search + python
Search Similarity Search

Pipeline – Custom python, Azure Langchain - customized, Custom Pipeline – Custom python, Langchain, Custom Python
Logic Apps Python Pipelines Pipelines

Page 73
Comparative Analysis – Enterprise Knowledge Management Pipelines

Option 1 (MS Azure Option 2A Python Option 2B Option 2C

Functionality pipeline, Azure FR, Redis Python pipeline+ Redis Cache, Python pipeline+ Redis Cache
Search+ Generative AI)
Cache, ADF + Generative AI ) Logic Apps + Generative AI + Generative AI

OOB Connectors for SharePoint, Has multiple built connectors ADF also provides OOB Logic App –alternative multiple No connector, custom
0365 and others for multiple systems multiple i/p connectors OOB Connectors implementation in python

RE-indexing of files, triggers & Reindexing, triggers possible Reindexing, triggers possible Not OOB, Custom Python code
Reindexing, triggers possible
Scheduler functionality (ADF) (Logic Apps) to be developed

Contextual Search Results – Med, uses Non GPT High, uses GPT embeddings High, uses GPT embeddings High, uses GPT embeddings
Accuracy of Results Embedding engine engine engine engine

MS Azure Search support Will need to leverage Codex for Will need to leverage Codex for Will need to leverage Codex for
More structured data Structured data Indexing, O- SQL query, alternate custom SQL query, alternate custom SQL query, alternate custom
DATA dynamic query O-DATA search code O-DATA search code O-DATA search code

Unstructured data (PDF, MS

Possible, OOB Cranking
Word, ppt, audio transcripts, Possible, Python pipeline Possible, Python pipeline Possible, Python pipeline
using in-built Pipeline
Images etc)

Custom (Custom implement), Custom (Custom implement), Custom (Custom implement),

Spell check, auto-fill OOB Feature
App level handling app level spell check App level spell check

Page 74
AI-Powered Auto KB Article generation for Shared Service/Contact Center

Problem Statement
 Auto Knowledge base article generation is typically a time-
consuming activity. Team Leader have to work with team to
identify what KB articles to be generated and have to involve
larger team to create the same

The Solution
 Open AI GPT3 state of the AI, can automatically read through the
call transcripts and help cluster corpus by issue type and then for
each issue type – can generate KB Articles automatically

Benefits
 This
can help save lot of manual work by Shared services /contact Open AI GPT 3 Provides very strong Generative Capabilities,
center teams witch helps in accurate KB Generation.

Page 75 Return back to meta data

GPT-3 Enabled Earnings Call Transcript Analysis -demo

Earning Calls
 Earnings calls are conference calls hosted by publicly traded companies to discuss
their financial performance for a given quarter or fiscal year.
 Financial
analyst needs to analyse these calls transcription to get understanding of
company’s financial growth and projection, margins, Inventory debts etc

Business Challenges
 Time-consuming: Manually analysing earning calls can be a time-consuming process,
as the analyst must listen to and transcribe the entire call, which can take several
hours
 Subjectivity: Manual analysis can be subjective,
 Data overload: Earning calls can contain a large amount of information, and manually
analysing them can lead to data overload, making it difficult to identify and prioritize
key insights.

How our Asset can help solve problem statement

 It enables analyst to find key insights and summaries of earning calls
 Analyst could ask query from earning call to get information about certain aspects

Page 76 Return back to meta data

ESG and Risk
Management

Return back to meta data

EcoPeer: An Industry GPT enabled ESG Supplier & Competitor Risk Analysis &
benchmarking Solution

EcoPeer empowers to evaluate ESG performance of an organisation against peers, ESG related recognitions or controversies and key ESG
insights from the company’s annual responsibility report
Tool Features
Comprehensive view of ESG performance of
suppliers or competitors

Strategize to improve sustainability

initiatives

Sustainable financing decision made easy

Identify potential reputational risks and

controversies of competitors

Better Management of ESG Risks

ESG Rating with ESG Initiatives ESG Benchmarking across peers

Improve your companies ESG footprint on

Social Media, helps improve ESG ratings

Tools & Technology Used

 GPT3  Huggingface  Power BI
 Yahoo Finance zeroshot  News API
classification
 ChatGPT  Social Media
pipeline
Social Media Controversies & Recognitions Annual Report Analysis

Page 78 Return back to meta data

Advanced Supplier & Legal Contract Analysis Tool

Problem statement
Companies relying on manual contract creation, contract review, and approval have high possibility to Benefits
overlook commitments, possible discounts, compliance mandates, and missing any hidden risks
 Reduced Turn Around Time in
Our solution Contract Management Lifecycle
An Advanced Supplier & Legal Contract Analysis tool which helps in the all the stages of Contract
 BetterDecision Making : With
Management Life cycle
valuable insights and suggested
negotiation strategy
 Improved visibility with clear and
actionable data/risks
 IdentifyNew Opportunity through
extracting key information from the
Quick Validation & Extraction of Ask Questions to Contact Document On the Go Checklist – Validates contract
key contract Data/Entities Type in a query to get a summary and find if key Clauses are present, flags
 Upfront
Risk Mitigation plans and
The tool extracts Legal entities Generative response from Contract document Validate a contract basis
and relevant clauses from along with the segment in the contract where commercial and Legal Checklist
Negotiation tips
Contracts on a click the data is present

Tools & technologies

 GPT3 API
 Blob

 Azure ML
Compare between contract clauses (standard Contract Risk Management –  Azure Cloud
and Vendor), Review differences, Identify against Controls
 Power Apps
induced Risks & provide Negotiation Tips

Page 79 Return back to meta data

SmARRF - Smart Auto Risk Register Filling

OpenAI GPT-3 large language models can be used for doing quick Analysis results of ERD Risk call transcript as is done by Risk Analysts and then this tool can
convert format and auto-fill Risk Register without much of AI training and can be game changer to deliver this quickly and cost-effective manner.

ERD Risk Call Recordings

Key benefits of leveraging
GPT3 provide following o/p large language models GPT3
Able to read though all RISK Able to go through
Pre- meeting transcripts to provide hundreds of meetings,
Processing Generative O/P and Generate Human
= Generates Key Risks Mentioned Like structured Output
along with Severity & Region using Generative AI
Unstructured Risk
Audio or teams call Can ask GPT3 to convert format 80% + Manual Efforts
transcript– ERD calls, to populate Risk Register too and Saving in analysis.
auto-fill, add Risk category, Reducing costs
speech Transcripts Severity, Risk Drivers etc significantly
Multiple other questions related to
Risk discussion can be asked ? Enhanced Insights

Prompt
Engineering

Page 80 Return back to meta data

LLM powered PEP & Background Verification and Due Diligence (WIP)

Problem Statement Background Verification Sources

 Existing
methods includes an Analysts going through documents, articles around internet, collecting
and preparing the document manually which is tedious and time and resource consuming task

The Solution
 The proposed due diligence solution offers a comprehensive solution to investigate potential
business partners, employees, or other entities. The tool collects data from various sources,
including public records, online databases, and social media platforms. It analyzes corporate
background, government touchpoints, litigation, criminal records, regulatory non-compliance,
violations, social media impact/abnormalities, and ESG factors
 Using GPT, the tool generates a concise summary report that provides users with a clear risk profile
of the entity under investigation. The report enables users to quickly assess the entity's overall risk
profile and make informed decisions based on comprehensive and reliable information

Benefits
 The solution will do the exact same but with efficiency, clear and concise reporting and
comprehensive coverage.
 The finance industry can use this tool to conduct due diligence on potential borrowers, investors, or
acquisition targets, ensuring compliance with regulatory requirements and mitigating financial risks.
 Law firms can use the tool to conduct due diligence on clients, opposing parties, or witnesses in
legal cases, ensuring compliance with legal requirements and mitigating legal risks
 HR departments can use the tool to conduct due diligence on job candidates, mitigating
employment risks

Page 81 Return back to meta data

TPRM Risk Management – Auto filling of Auditor Query responses from Vendor SOC
Reports

Problem Statement 01 02 03
 During TPRM Audit, typically Auditors would put forward lot of Get Inputs Data Cracking Generative Respons
queries related to key risks, if say your application is running on Normalization extracted from
File with queries
Cloud vendor and 3rd party Enterprise Vendors software . Finding Data Normalization of queries Documents
response to these queries is tedious job and requires people to Vendor Documents
Use GPT Models Use Combination of
read through multiple SOC 1, SOC 2 reports and documents to Automation and ML
draft responses
Auditor Queries Vendor SOC Reports

The Solution
 The proposed GTP Enables solution is able to understand Auditor
queries, provided as CSV, or unstructured text, and then analyse
SOC Reports (SOC 1, SOC 2 reports) to help automatically
populate responses to auditor queries from these Vendor SOC
documents (Azure SOC Reports, AWS SOC Reports, SAP etc)
 It can analyse multiple vendor SOC reports at scale to find out
Responses to Auditor queries with respect to compliance /controls

Benefits
Generative Auditor
 Timeend efforts saving for Audit work, Automation of task with queries Responses O/P
minimal errors

Page 82 Return back to meta data

Social Media
Analytics & Hyper-
personalization

Return back to meta data

Product Catalogue Description Generation Using GPT-3

Problem Statement
 Retailer/merchandise has to manually create/update all description and
title for each product
 It could become time-consuming and cumbersome process because of
large category of products
Business Challenges
 Content Management: Creating and managing content for the online
catalogue can be a time-consuming task.
 Competitive Analysis: Understanding the competition and staying up-
to-date with industry trends is important for maintaining a competitive
advantage.
 User Experience: Creating an online catalogue that is user-friendly and
easy to navigate is essential for engaging and retaining customers.

The Solution
 The solution enable retailers to upload their product image
 Retailer can generate product title and product description using GPT-3
 Retailer can provide some persona to generate the description specific
to certain customer segments.

Page 84 Return back to meta data

Social Media LLM Content Analysis

Select Social
Media

Specify Search <Specify the Search term e.g. #ChatGpt>

Item

Specify timeline Last 5 hours Last 24 hours Last 1 week

XX XX XX
Key Stats Total number of posts Total number of Total Traction Generated (Sum of total
verified posts no. of views of all post combined)

<this section will be populated by the output generated by ChatGPT

Summary
by click of button Generate Summary below>

Generate Summary

Page 85 Return back to meta data

Hyper personalisation use case using psycho –linguistic Customer behaviour analysis
using GPT-3 & Deep Learning models
WIP Demo Ready
Customer sees a
Utilizing abundantly available personalized view of NBT Model predicts Web crawler service
unstructured text data & sequential CC Customer clicks on the next best scraps customer
Customer logs his shopping ‘Nearby offers’ tab to transaction of Social Media content
tranx data as input to GPT-3 powered in his mobile trends/dashboard, see personalized customer[ Merchant (Twitter,FB,
platform along with deep learning banking app CC tranx. offers category] Instagram, Google)
capabilities that ingests and aggregates
1 2 3 4 5
customer data in real-time, provide
tailored experiences utilizing behavioral
science techniques and continuously
improve by capturing the engagement
efficacy and conversion results.

10 9 8 7 6
Value Generated Invoke Nudge ML Utilize Consumer GPT-3 recommends GPT-3 LLM ranks Utilize GPT-3 LLM
models to send offer persona and predicted products from customer’s to perform Psycho-
results to curate previously predicted identified linguistic analysis
Banks to customer on personalized category as per personality traits on extracted text
preferred engagement promotional offer for identified on scale of 1-10
 Generate revenue through Merchant tie-
channel customer personality trait
ups/cross selling options
Merchants Key Resources Used GPT generated Offers
 Offer creation: working closely with
merchants to produce automated and
Social Media
intelligent campaigns built from descriptive Footprint
GPT based
insights Merchant-
Consumer
Product
Customer needs and
Catalog GPT-3 based trait
behaviour
 Personalized offer dashboard identification

Page 86 Return back to meta data

Healthcare
Generative AI
Use cases

Return back to meta data

Claims Denial Rebuttal

Problem Statement
 Insurance claim denials can be a frustrating and time-consuming process for doctors, who often have to
deal with complex medical jargon and insurance policies.
Prompt
Response:

Proposed Solution
 GPT-3 can assist doctors in drafting a persuasive rebuttal email by providing relevant references,
analysing insurance policies, and suggesting appropriate language to use. For example, GPT-3 can
analyse the language used in the insurance company's denial letter and provide suggestions for how to
effectively refute the claims made by the company.

Benefits
 Saves time
 Provides structure and guidance
 Provides access to a vast database of knowledge
 Increases the chances of success in convincing the insurance company to reconsider the claim denial can
be increased.
 Improved efficiency

 Enhanced persuasiveness

Page 88 Return back to meta data

Claim Denial Email

Problem Statement
 Insurance claim denial emails can be a frustrating and time-consuming process

Proposed Solution
 The ChatGPT can help in creating email draft for every claim which is denied by the payer. For an
example, The ChatGPT help create the claim denial emails for all claims denied depending on rejection
rules

Benefits
 Saves time
 Create customised email depending on rejection rules
 Accelerate the process
 Future automation could be done on single click the email send to the provider and patient

Page 89 Return back to meta data

MedMind AI

Problem Statement
 Existing methods for prescription and test interpretation often rely on patients' ability to understand
medical jargon. This can be difficult for non-medical professionals, leading to misunderstandings and
errors. The solution can improve on these methods by providing patients with a clear and concise
explanation of their medications and tests, reducing the chances of errors and misunderstandings.

Proposed Solution
 The proposed solution is a specialized AI system that can read and interpret medication and lab test
instructions from prescriptions or medical recommendations. It can provide users with information such
as the medication's name, dosage, usage instructions, and side effects, as well as information on test
pre-requisites, such as fasting requirements or the need for other medical tests for undergoing specific
tests (GPT-based Assistance for medications and lab test information)

Benefits
 Improving patient compliance, Enhancing patient understanding
 Reducing healthcare professional workload
 The solution can be used in a wide range of healthcare settings, making it a versatile and scalable
solution. It can benefit patients with chronic conditions who require long-term medication management,
as well as those who receive frequent test recommendations.
 This will be promising solution that has the potential to transform medication and test interpretation in
healthcare, improving patient outcomes and enhancing the efficiency of healthcare services.
 This can lead to improved patient outcomes and reduced healthcare costs, as patients are less likely to
experience medication errors or require follow-up appointments to clarify medication or test instructions.

Page 90 Return back to meta data

DoctorAssist

Problem Statement
 Consultation with healthcare professionals who can perform physical examination and order appropriate
test based on symptoms and Healthcare org such as WHO or NMA develop clinic guidelines for diagnosis
and treatment of various diseases might not always be up to date

Proposed Solution
 GPT-based Assistance for Disease diagnostics : The proposed solution assists doctor in evaluation of a
specific disease. It can help doctors in getting deep knowledge regarding the disease such as SOAP,
Recommended test, etc.
 Medical Research: AI can assist researchers in identifying promising avenues for further researches and
development
 Clinical Decision making: AI can help healthcare professionals make more informed decisions
 Medical Education: Aspiring healthcare professionals can benefit from learning from AI.

Benefits
 Improved awareness and understanding: Many doctors may not be aware of the various diagnostic tests
available for a particular disease or may not fully understand what each test entails. By providing clear
and concise info on these test, it can help make informed decisions about the disease. Access to up-to-
date model.
 .Reduced healthcare cost: Appropriate diagnostic testing can lead to more targeted and effective
treatment, thereby reducing healthcare costs.

Page 91 Return back to meta data

HR Generative AI Use
cases

Return back to meta data

Applicant Resume Screening and CV to JD match using Generative AI

Background The Solution

 Recruiting and hiring processes involve the evaluation of numerous  A Generative AI tool which is capable of screening the CV based on
resumes received for each job opening. Manual resume screening is time- Job description or specific skillset
consuming and prone to human bias, leading to inefficiencies and potential
oversight of suitable candidates  Uses Open AI to match the resumes and provide the best fit
candidates
 It also provides a summary and points relevant in the resume for
which the candidate has been selected for specified JD/skillset
requested to the tool
Benefits
 Increased Efficiency: it reduce the time spent on resume screening
 Improved Candidate Evaluation: The automated screening accurately
identifies highly qualified candidates, reducing the likelihood of
overlooking suitable candidates
 Minimized Bias: It provides a fair and unbiased evaluation of candidates,
minimizing human bias in the screening process

Tools & Technologies

 MS Open AI GPT 4  Blob

 Azure ML  MS Form recogniser
 Azure Cloud

Page 93 20 August 2024 Return back to meta data

Voice of Employee Analysis using Generative AI

Background The Solution

 Employee Performance Reviews, Employee Feedback , Glassdoor and Exit  VOE provides Employee level Insights viz. Employee sentiment,
Interview’s provide excellent untapped Resource for getting Employee Satisfaction Index, Employee’s strength and support required
emotional Insights which can eventually help to identify opportunities to
improve HR Services and increase Employee satisfaction and retention  It is also designed to show a Manager level Insights for driving a
better team

Benefits
 Better Employee Engagement and support with Productive workforce
 Increase Employee Satisfaction in an Organization
 Reduce Attrition Rate by unearthing hidden soft attributes

Tools & Technologies

 MS Open AI GPT 4  Blob

 Azure ML  Power BI
 Azure Cloud

Page 94 20 August 2024 Return back to meta data

GenAI
Use Case Library
GtM Use Case Repository

Last update date: 22.11.2023

Supplier negotiation cockpit (1/1)

Negotiation cockpit consolidates scattered and unstructured supplier information and

generates negotiation strategy and agenda via diagnostic and predictive AI
1 Case example: Supplier negotiations cockpit USE CASE | ILLUSTRATIVE

Supplier negotiations in category management Deep dive – Negotiation cockpit

Laborious and time intensive process of supplier

negotiation preparation due to distribution of
“ Prepare comparison of similar contracts to that of Supplier A and
prepare a negotiation agenda.”

Problem
€
Contract conditions Inventory Commercial External factors
contract and product performance information over ► Stock levels ► Shelf turnover
► Product, price & quantity ► Raw material costs
many unstructured sources ► Payment ► Inventory rate ► Competition price /
► Delivery & buy-back turnover rate ► Profit margins product
► Duration, termination & renewal ► … ► Backlog rate architecture
► Quality & guarantee ► Quality and ► News about
► Service level satisfaction industry and the
► Marketing and sales subsidies ► … suppliers
► Kick-back terms ► …
Data processing and standardization
► …
Use AI capabilities to consolidate supplier and Information extraction from
contract documents possible
procurement relevant data from different sources in Embedded semantics vectorization
Solution Hierarchical structure based on Optimization metrics
different formats and generate a negotiation agenda supplier and category level Optimization criteria:
and strategy ► Gross margin
► Net margin (w/ kick-back)
Diagnostics AI Predictive AI ► Payment conditions
► How was historical supplier port- ► Normal vs. product scale
► Define buy-in price corridor
folio sales volume and margin? conditions
► Which kick-back condition
► How was historical service level AI ► Old goods investments
maximizes net margins?
performance? ► Marketing subsidiaries
► Which payment method is
► Which contract terms were
most suitable for WC? Supplier KPIs:
► Minimization of manual work during supplier evoked?
► On-time delivery reliability
Realized ► …
► Order accuracy
contractualization
business ► Improved buy-in prices leading to margin gains
► Lead times
► …
impact Negotiation cockpit Outlook
► Increased bargaining power against suppliers Standardized briefing and negotiation support with supplier Monitor negotiation process and
bonus, purchase price reduction, cash contribution, kick-back performance benchmarked to targets
conditions but still leaving room to address individualities

Source: EY-Parthenon project expertise

Return back to meta data

Customer negotiation cockpit (1/1)

Negotiation cockpit consolidates scattered and unstructured customer information and

improves the negotiation preparation with analytical AI
2 Case example: Customer negotiation cockpit USE CASE | ILLUSTRATIVE

Customer negotiations in Sales & Key Account Deep dive – Negotiation cockpit

Contract view Historical view External view

► Product, price & quantity ► Past negotiation ► Raw material prices
Complex and time intensive process of negotiation ► Fees & drift outcomes ► Macroeconomics
preparation with customers combining information ► Counterparts (promotions etc.) ► Former negotiation trends and industry
Problem ► Payment terms & conditions approaches and news
such as business plan, product performance, changes ► Delivery & logistics strategy ► …
► … ► …
in demand of fees, drift, terms and conditions, etc.

Extracting, standardizing GPT context abilities Q&A chatbot can assist by

and processing data with understanding and putting answering all contract
the help of AI information in context relates requests

Use AI capabilities to consolidate contract data,

historical performance and external factors from
Solution
multiple sources in multiple formats and create a Automized contract analysis Predictive AI
negotiation agenda and strategy in a single cockpit ► How was historical sales volume and margin with customer? ► Define upfront fee corridor
► How did fees and drifts develop with customer? ► Which price per quantity
► How were historical terms & conditions and their execution? maximizes net margins?
► Which regions, products and outlets were included? ► Which payment terms are most
► Which contract terms were neglected? suitable?
► … ► …
► Minimization of manual work during customer /
retailer negotiation phase
Realized
► Increased negotiation power through
business transparency
Negotiation cockpit with all relevant customer information making
impact negotiation preparation much faster and more convenient allowing a
► Improved contract conditions (prices, drift, better positioning towards the customer
terms, …)

Source: EY-Parthenon project expertise

Supplier contract analysis (1/2) Use case category:
L4: Implemented with client
Deploying ChatGPT with supplier contracts enables AI-based support for the L3: Built for EY

day-to-day work of every CM and automated supplier contract analysis

L2: Built as a Prototype
L1: Proof of concept

3 Case example: Supplier contract analysis USE CASE | ILLUSTRATIVE

Supplier contract analysis in category management POC capabilities

Reads & consolidates +500 supplier contracts

Supplier negotiations for better buy-in prices are A
labor and time intensive, due to scattered and
Problem unstructured data from [Client]’s complex supplier
landscape GPT context abilities understand complex inputs and
B
put information in context

Q&A chatbot will assist CMs by answering all supplier

Use AI-capabilities to process & aggregate supplier
C contract related requests
contracts, product offerings and sell-in amounts in
Solution dynamic databases and then leverage Generative AI
to provide relevant information to CMs for Automated contract analysis enables contract
optimization of procurement conditions
D
comparison and consolidation

AI-generated supplier briefings will greatly facilitate

► Minimize manual work E and standardize preparation of supplier negotiations
► Improved contract terms (buy-in process) and with dynamic tools
Realized
thus margin gains
business
► Increased bargaining power through improved Negotiation briefings give dealmakers the most
impact transparency F relevant information for upcoming talks in a concise
► Fully informed, fact based decisions output format and on-demand
Deep dive available
Source: EY-Parthenon project expertise

Return back to meta data

Supplier contract analysis (2/2) Use case category:

Some modules of Supplier Contract Analyzer are openly available while

L4: Implemented with client
L3: Built for EY

others containing specific business requirements must be coded individually L2: Built as a Prototype
L1: Proof of concept

3 Schematics of Supplier Contract Analyzer application ILLUSTRATIVE

Standard components are available “off the shelf” &

don’t need adaptations

Text
Document Information User
to
reader processing interface
text
Business Manage
& &
LLM
prompting retrieve
PDF Dataset logic information XLS
and GPT-4
output model
handling
Input Output

GPT made working with text and language

very easy compared to traditional
complicated and unreliable NLP1 solutions
Application Text
parameter EY-P value-add on proprietary to
handling XLS
components with tailored integration
of business expertise
Standard modules Proprietary modules
1.Natural language processing
Source: EY-Parthenon

Return back to meta data

Customer contract analysis (1/2)

Deploying ChatGPT with contracts enables AI-based support for the day-to-day work of
every salesperson, customer manager and automated contract analysis
4 Case example: Customer contract analysis USE CASE | ILLUSTRATIVE

Customer contract analysis in sales POC capabilities

Reads & consolidates +500 customer contracts

A
Customer negotiations are labor and time intensive
Problem due to scattered and unstructured information
sources
GPT context abilities understanding and putting
B
information in context

Q&A chatbot will assist every salesperson by answering

C all customer contract related requests
Use AI-capabilities to process & aggregate customer
Solution contracts and leverage generative AI to consolidate
information in contract Automated contract analysis enables contract
D
comparison and consolidation

AI-generated supplier briefings will greatly facilitate

 Minimize manual work E and standardize preparation of customer negotiations
Realized  Improved contract terms (sell-in process) and
thus margin gains GDS AI
business  Increased bargaining power through improved Negotiation briefings 4 page contract 35 page contract
impact transparency F 3h per contract 10min per contract
 Fully informed, fact based decisions
€50-60 €3
Deep dive available
Source: EY-Parthenon project expertise

Source: EY-Parthenon project expertise Return back to meta data

Customer contract analysis (2/2) Use case category:

Some modules of Customer Contract Analyzer are openly available while

L4: Implemented with client
L3: Built for EY

others containing specific business requirements must be coded individually L2: Built as a Prototype
L1: Proof of concept

4 Schematics of Customer Contract Analyzer application USE CASE | ILLUSTRATIVE

Standard components are available “off the shelf” &

don’t need adaptations

Text
Document Information User
to
reader processing interface
text
Business Manage
& &
LLM
prompting retrieve
PDF Dataset logic information XLS
and GPT-4
output model
handling
Input Output

GPT made working with text and language

very easy compared to traditional complicated
and unreliable NLP1 solutions
Application Text
parameter EY-P value-add on proprietary to
handling XLS
components with tailored
integration of business expertise
Standard modules Proprietary modules
1.Natural language processing
Source: EY-Parthenon

Source: EY-Parthenon
Assistant for contract management (1/1)

AI-driven [lease agreement] support offers benefits such as automated document

analysis, improved compliance and risk assessment as well as portfolio management
5 Case example: Assistant for contract management

Contract assistant Deep dive – Contract / Lease agreement assistant

Internal data External data

Contract management in the real estate sector is Existing contracts and lease Local laws and changes in
Problem highly complex due to unique contracts and lease
agreements
Client/ Sector specifics
agreement
Lease performance
regulation
Changes in market
(behind the placeholder there is a content relevant
criteria for Real Estate/
environment
Owners meeting Construction sector)Regional, location specific
notes nuances

An AI-powered ‘contract assistant’ can effectively

optimize work by identifying crucial factors such as
Solution
expiring contracts and providing guidance on required Language model combines … providing a single source
LLM
actions to be taken own capabilities with of truth for contract and
internal and external data, … lease agreement related
requests

Realized
► Efficient contract and portfolio management
business Contract management assistant Contract writing assistant
► Improved compliance and risk assessments
impact ► Tracking of upcoming contract expiration ► Preparation of next negotiation rounds
and renewal dates and initial drafting of contracts
► Providing an action plan for contract- ► Identification of potential areas for
related events, such as expiration improvement based on similar contracts
Solutions OpenAI ► ... ► ...

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Automated supplier inquiry handling (1/2)

Parts of supplier interaction like payment status inquiries trigger manual effort across
different departments – AI is able to automate supplier interaction to reduce costs
6 Case example: Automated supplier inquiry handling USE CASE | ILLUSTRATIVE

Extended process automation Deep dive – automated supplier inquiry handling

Supplier inquiries (e.g. regarding unpaid invoices) are

Problem often handled manually via supplier management or Inquiries from the supplier are received via
helpdesk with high manual effort e-mail, chat-function in the supplier system
or by telephone (voice bot)

Existing
Generative AI helps through automated inquiry data, e.g.:
► ERP
Solution management interacting with different systems (e.g.
► CRM
finance systems AP) using documented information Data is processed and compared with existing
► Account

Target Process with

company information, taking into account the
country and specific requirements of the inquiry Payable

Generative AI
► Process efficiency by reducing human
Realized interaction with respective systems / customer
business satisfaction by increased speed in inquiry
impact handling
► Process cost savings The sub-process is initiated and processed using a
software robot (e.g. an RPA bot) and the supplier
is automatically notified of his request Supplier
response
Solutions OpenAI

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Streamlining the handling of incoming supplier requests through automated processing
– increased degree of automation possible across all process steps
6 Deep dive: Automated supplier inquiry handling DEEP DIVE | ILLUSTRATIVE

Process start Process Input Request clarification Request processing Process end

As-Is Process
Sales

QM
AP

The inquiry is recorded by an The responsible persons such as The steps for processing the
Inquiry from Supplier response
employee with knowledge of the Quality Management, Sales, and request is initiated and the
supplier about about status of
respective language and Accounts Payable are being supplier is informed about the
unpaid invoices is request is send out
maintained in the ERP system contacted for further information status of the request
made

Degree of automation
with RPA

Target Process with Generative AI

Existing data
(e.g., ERP, CRM, AP)

Inquiry from Inquiries from the supplier are Data is processed and compared The sub-process is initiated and Supplier response
supplier about received via e-mail, chat-function with existing company processed using a software robot about status of
unpaid invoices is in the supplier system or by information, taking into account (e.g. an RPA bot) and the supplier request is send out
made telephone (voice bot) the country and specific is automatically notified of his
requirements of the inquiry request
Degree of automation
with AI
Automated customer inquiry handling (3/3)

Parts of customer interaction trigger manual effort across different departments –

AI is able to automate parts of customer management to reduce costs
7 Case example: Automated customer inquiry handling USE CASE | ILLUSTRATIVE

Extended process automation Deep dive – automated customer inquiry handling

Customer inquiries (e.g. regarding invoices) are

Problem often handled manually via sales team or helpdesk Inquiries from the customer are received via
with high manual effort e-mail, chat-function or by telephone (voice
bot)

Existing
Generative AI helps through automated inquiry data, e.g.:
► ERP
Solution management interacting with different systems (e.g.
► CRM
finance systems AR) using documented information Data is processed and compared with existing
► AR

Target Process with

company information, taking into account the
country and specific requirements of the inquiry

Generative AI
► Process efficiency by reducing human
Realized interaction with respective systems / customer
business satisfaction by increased speed in inquiry
impact handling
► Process cost savings The sub-process is initiated and processed using a
software robot (e.g. an RPA bot) and the
customer is automatically notified of his request Customer
response
Solutions OpenAI

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Automated customer inquiry handling (2/3)

Streamlining the handling of incoming customer requests through automated

processing – increased degree of automation possible across all process steps
7 Deep dive: Automated customer inquiry handling USE CASE | ILLUSTRATIVE

Process start Process Input Request clarification Request processing Process end

As-Is Process
Sales

QM
AP

The inquiry is recorded by an The responsible persons such as The steps for processing the
Inquiry from Customer
employee with knowledge of the Quality Management, Sales, and request is initiated and the
customer about response is
respective language and Accounts Receivable are being customer is informed about the
new invoices is send out
maintained in the ERP system contacted for further information status of the request
made

Degree of automation
with RPA

Target Process with Generative AI

Existing data
(e.g., ERP, CRM, AP)

Inquiry from Inquiries from the customer are Data is processed and compared The sub-process is initiated and Customer
customer about received via e-mail, chat-function with existing company processed using a software robot response is
new invoices is or by telephone (voice bot) information, taking into account (e.g. an RPA bot) and the customer send out
made the country and specific is automatically notified
requirements of the inquiry
Degree of automation
with AI
Automated customer inquiry handling (3/3)

Parts of insurer interaction like request for reparation imbursement trigger manual
efforts of car specialists – AI is able to automate insurer interaction to reduce costs
7 Case example: Automated insurance inquiry handling USE CASE | ILLUSTRATIVE

Extended process automation Deep dive – automated insurance inquiry handling

Insurer inquiries (e.g. regarding non partner garages

Problem quotes) are often handled manually via car specialists Damage documents from the garage are
who check line by line with high manual effort received via e-mail, with line items that
need to be repaired

Existing
Generative AI helps through automated inquiry data, e.g.:
management interacting with different systems (e.g. ► ERP
Solution
ERPs & databases) comparing prices and alerting ► CRM
Data is processed and compared with existing
discrepancies ► Database

Target Process with

reparation information, taking into account the
specific requirements of the inquiry

Generative AI
► Process efficiency by reducing human
Realized interaction with respective tasks / customer
business satisfaction by increased speed in inquiry
impact handling
► Process cost savings The sub-process is initiated and the insurance is
automatically notified if there are discrepancies
in this request Insurance
response
Solutions OpenAI

Source: Desk research; EY-Parthenon project expertise

Source: Desk research; EY-Parthenon project expertise

Product data quality (1/1)

AI driven quality mechanism continuously ensures accurate and comprehensive product

database through clearly defined requirements, issue mitigation approach and validation
8 Case example: Product data quality USE CASE | ILLUSTRATIVE
Data quality guardrails definition Data quality issue detection Exemplary product issue & fix
Product Material Weight Diameter ► Definition of minimal required ► AI crawls through existing
data quality standards and Reference: Full info Flagged issue: Limited info
Fan A
Alumi-
num
2.5kg 137cm databases
guidelines
► AI identifies products with
Fan B Steel ? ? ► Systemic definition of rules and
missing data, based on a lack
possible data quality issues (e.g.
... ... ... ... of typical product data for
missing field, duplicates)
that category

1
2
GenAI supported info retrieval

Validation and monitoring AI driven ► LLM then extracts contextual

Client specific
data quality Functional specifics missing
5 product information from the e.g. product size dimensions,
► Data quality is validated improvement manufacturer website operating voltage in volts,
regularly 3 ► AI can scan additional sources
protection class, remote
control
► AI scans for recurring issue to enhance data pool (e.g.
patterns and triggers manual 4 product PDFs, contracts) Solution: scan data from other available information sources
investigation, if needed such as the manufacturer website and extract info

GenAI supported data correction

► GenAI identifies the relevant fields in product database to insert newly Functional
extracted information specifications

► Automated crawler systematically fills and completes the product data

master

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

AI-powered database accumulates company-wide knowledge and makes it available to
everyone by interaction with a chatbot – without the need of manual document curation
9 Case example: Knowledge database USE CASE | ILLUSTRATIVE

Knowledge Database powered by GenAI Deep dive – Knowledge database set-up

Web frontend Custom backend

With increasing scale and formats, current Unification of data formats

Question / to be usable for LLM
Problem approaches to knowledge management tend to grow
Instruction
in complexity, making it unintuitive to use
“How can I solve … ?
Internal documents
Interact with and data suitable for .pdf
underlying documents AI searches and .pptx
retrieves most use case, e.g., .doc
by chatting product descriptions
relevant data HTML
AI database that allows knowledge extraction with or manuals
Solution proprietary documents through language models
without manual document pre-processing
Client
specific LLM
Language model combines own
knowledge from training with the
retrieved internal data, returning
an answer that is relevant to the
► Improved productivity for research tasks
Realized user
► Elimination of need for extensive manual
business knowledge management Data security ensured:
impact ► Service improvement for customer knowledge DB Storage on internal (EY) or
Answer incl. citations
company servers –
and sources – all Response No data kept by LLM or
based on own data used for other purposes
Solutions OpenAI

Source: EY-Parthenon project expertise

Return back to meta data

Customer onboarding assistant (1/1)

GenAI powered customer onboarding assistant use natural language processing to

solve customer problems in an efficient way while creating conversational experience
10 Case example: Customer onboarding assistant USE CASE | ILLUSTRATIVE

Customer onboarding assistant How it works?

Web frontend Custom backend
► Time-consuming and semi-automated customer
onboarding process to platform services, leading
Customer
Problem to waiting times and customer care efforts
language input
► Unguided onboarding process statistically leads
“How can I help you … ?
to higher churn rate Unstructured customer
Process any request of Language trans- data transformed in
the customer and formed in GenAI
processable machine analyzable data
research accordingly in various formats such
prompts
► AI database that allows knowledge extraction as text and voice
Solution with proprietary documents through language
models without manual document pre-processing
GenAI refers to the knowledge
GenAI base for customer inquiries by
leveraging onboarding documents
and videos
► Improved productivity for research tasks
Realized
► Elimination of need for extensive manual
business knowledge management Problem solution in a
impact ► Service improvement for customer knowledge DB natural, accurate, and
Provide customer
context-aware manner that
with profound Response further creates positive
response and conver- customer experience
Solutions OpenAI sational experience

Source: EY-Parthenon project expertise

Source: EY-Parthenon project expertise Return back to meta data

Automated ad creation (1/1)

[Client] can introduce an ‘AI automated ad creation’ feature in its core product to
automate personalized generation of content for ad copies and features
11 Case example: Automated ad creation USE CASE | ILLUSTRATIVE

Automated ad creation How it works?

Web frontend Custom backend
► Advertisers invest significant resources in
Customer Customer data
creating content for ad copies, visuals, and
content input collection
Problem creative elements
► Generating optimized content remains a time-
consuming and challenging task Customer data i.e.
Upload and/or prompt Natural Language Processing demographics and
(NLP) capabilities are used to website visits,
for ad campaign analyze the customer’s
content content database
► Analyze customer preferences, interests, and language, tone, syntax
themes, individual
behavioral metrics
content and third
Solution ► Copywriting - GenAI powered natural language
party data
feature can assist in creating personalized
content for ad copies GenAI processes input and
GenAI standardizes it to LLM for automated
ad creation. Standardization ensures
generated content is in accordance
► Efficiency increase of ad campaign creation
Realized with LLM for consistency and quality.
► Create customized content that resonates better
business
with the target audience and provides enhanced Provide customer with Ad campaign delivery in
impact engagement profound response to most convenient format for
inquiries, questions the customer (e.g. multi
Response modal format like textual,
and feedback quickly
visual, or voice)
Solutions OpenAI and accurately.

Source: EY-Parthenon project expertise

Source: EY-Parthenon project expertise Return back to meta data

Sales content creation (1/1)

GenAI can help [Client] automate draft personalized content for proposals and
contracts, besides calculating prices, discounts and taxes in sales quotes
12 Use case example: Sales content creation USE CASE | ILLUSTRATIVE

Sales content creation How it works?

Process proposal/marketing
► Sales teams manually draft marketing brochures brochure creation request of
or proposals by coping & pasting info into text or the customer
presentation documents
Problem “Please provide … ?
► High volume of small and mid-sized clients
leading to increased efforts in marketing and Load of desired Load of additional data
sales information context for personalization

Language transformed
► GenAI analyze predefined templates & insert in GenAI processable ► Cookie analysis and click
gathered data in appropriate sections of prompts and adding behavior
marketing or sales proposal material relevant tags and ► Pricing data
Solution ► GenAI create content for marketing or sales metadata to content ► Existing and historical offers
proposal material
and marketing material
► GenAI creates personalized quotes based on
GenAI ► Customer relationship data
historical data Customer classification
proposal; (Gen)AI drafts
marketing materials
► Increased marketing and sales process efficiency based on templates,
Realized by scaling automated customer targeting pricing proposal based
on the customer's
business approach
► Increased customer conversion rate by scaling
classification and Sales content creation
impact historical data, budget
personalized sales and marketing material with personalized
and requirements content and pricing

Source: EY-Parthenon project expertise Return back to meta data

Customer service co-pilot (1/1)
GenAI can help [Client] provide transformative omni-channel customer engagement via real-time
sentiment analysis of user’s call and recommend resolutions to customer issues in real-time
13 Case example: Customer service co-pilot USE CASE | ILLUSTRATIVE

Customer service co-pilot How it works?

Web frontend Custom backend
Customer
► Customer service teams manually analyze
language/text input
Problem customer data, or are short of enough
information to respond to customers queries
“How can I help you … ?
Unstructured customer
Process any request of Language trans-
formed in GenAI data transformed in
the customer and processable machine analyzable
research accordingly prompts data in various formats
► AI driven inference analysis - AI engines can
extract insights from social media, customer such as text and voice
interactions, perform sentiment analysis of user‘s
Solution call & recommend resolutions to issues faced by
customers within a single click, by scanning GenAI refers to the knowledge base
through company sources, relevant knowledge for customer inquiries by leveraging
GenAI
articles1 CRM, ERP and contract data

Problem solution in a
natural, accurate, and
Realized context-aware manner that
► Reduced issue resolution time Provide customer
business that supports the customer
► Better quality of customer service agent with profound care agent’s problem
impact response and conver-
Response
resolution with
sational advice recommendation

Source: EY-Parthenon project expertise Return back to meta data

Intelligent Payroll Chatbot (1/1)

GPT-4 Powered Chatbot for Microsoft Payroll inquiries

14 Case example: Intelligent Payroll Chatbot USE CASE | ILLUSTRATIVE

Intelligent Payroll Chatbot Deep dive – Knowledge database set-up

Web frontend Custom backend
Due to HR resources constraint manually handling Employees can Ingest:
inquire the chatbot 1. Country-specific payroll topics
numerous inquiries from employees on payroll and HR
Problem on any topics related 2. General payslip structure
topics, there are delays or even instances where
to data ingested by 3. Company specific payroll
questions remain unanswered policies/FAQs
the bot. Chatbot
experience currently 4. Connect to relevant systems
entails:
EY developed a chatbot-based proof-of-concept Train chatbot:
(PoC), leveraging Azure OpenAI’s GPT APIs, for EY Multi-Language 1. Data ingestion
Capability 2. Logical chunking and retrieval
Solution and Microsoft’s Global Payroll Operate (Tax) to
User ability to augmentation approach
streamline the process for employees to receive
give feedback 3. Prompt engineering
responses to their payroll-related inquiries.
in UI
Human-in-the-loop incorporated through a
► Accelerated response times: no more delays in Increased trust
through metric dashboard to drive actionable
responses due to limited human resources
Confidence Score improvements to the chatbot
Realized ► Ability to address all questions: no inquiries getting
and Source ► Chat history of user feedback by source
lost as all are immediately answered and tracked
business through chatbot history Document Viewer documentation
impact ► High accuracy demonstrated: 92% accuracy rate
Highly accurate
► Summary of user ratings and written
► Optimize employee resources: human resources are responses feedback
now able to utilize time for more strategic work through ► Resolution rates
intelligent ► Response times
Solutions Azure OpenAI Azure Cache Redis (Vector DB)
context retrieval ► User volume metrics
Azure Blob Storage Azure AD

Source: EY-Parthenon project expertise Return back to meta data

Sales force assistant (1/3) Use case category:

Digital sales assistants can structure and analyze customer data and form
L4: Implemented with client1
L3: Built for EY

recomm. based on personal data, shopping behavior, and other patterns L2: Built as a Prototype
L1: Proof of concept

15 Case study: Sales force assistant USE CASE | ILLUSTRATIVE

Sales capability enhancement and support Deep-dive: AI-driven sales force assistant

2 Based on previous Sales rep 3 Decision of sales rep

customer relationship triggers the
Current customer management in sales relies on knowledge, sales rep preparation of new
manual tasks and personal experience – multiple decides on best fitting offering text/document User
Problem time consuming analysis loops with BI department offer / option for their in the same form of
interface
account historical documents
are required to create data-driven sales approaches layer

1 Q&A chatbot provides

responses with various
offering options for LLM GenAI layer
Combine power of traditional and generative AI to each client
create offering alternatives based on historical
Solution patterns and various success measures which can
be cherry picked by sales representatives Traditional
Multi-layer neural network as a conventional ML algorithm to conduct pre-
defined analyses to tailor offers AI layer

Step1: Recognize Step 2: Produce Step 3: Develop offer

purchase patterns similarity matrix among options based on pre- Business
similar accounts defined criteria logic layer
► New business opportunities (e.g. subscription
Realized
recommendations based on purchasing patterns) Basket size, share
business Competitive
► Increased sales force effectiveness through of wallet, basket
impact better sales engagement preparation
advantage of mix, etc. Input layer
METRO
Transactional data Historical offers

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Sales force assistant (2/3)

Digital sales assistants can structure and analyze customer data and form meaningful
recommendations increasing the sales force efficiency
15 Case study: Sales force assistant USE CASE | ILLUSTRATIVE

Automated customer briefings Individualized customer offers

► Generates individualized sales ► Provides customer tailored
briefing documents per customer offerings to sales reps via chatbot
► Provides concrete upselling ► Generates conditions in context
suggestions based on reference based on interaction with sales rep
basket in the customer category ► Automizes offer letter generation
and personalization, increasing
► Today, [Client]’s existing data wealth sales efficiency
is it not fully leveraged – often Deep-dive
limited by manual analysis efforts

► Generative AI can increase sales Sales rep

force efficiency by best supporting support
customer work in multiple was
Sales negotiation chatbot New customer acquisition
► Supports sales reps to negotiate ► Identifies new customer leads via
better conditions with customers AI-driven geo-analyses
► Allows alternative simulation of ► Supports first-touchpoint
offer conditions during meetings preparation by automatically
► Offers data-backed reasoning creating tailored sales pitches
support for ad-hoc questions based on public data

Source: Desk research; EY-Parthenon project expertise

Sales force assistant (3/3)

Making the GenAI sales force assistant available to sales reps enables efficient creation
of customer tailored offers and terms, slimming down labor-heavy preparation processes
15 Case study: Sales force assistant USE CASE | ILLUSTRATIVE

Sales rep… AI model(s)…

…analyzes historic customer data and scans for relevant insights

…selects desired customer in the sales tool to start the process based on pre-defined business logic (e.g. customer segment,
1
and gets a quick overview of key facts based on historic data share of wallet, category penetration)

…draws on pre-defined options (e.g., trial discounts, etc.) and

…queries the interactive chatbot to receive individualized sales
2 highlights margin potential based on spending and profiles of
strategies and actions tailored to customer
similar customers

…calibrates contract terms and decides on boundary conditions …provides a draft offer, including margin impact analyses
3
suggested by the AI (minimum amounts, timeframes, etc.) forecasts based on category heuristics

…receives offer draft and conducts fine tuning before sharing it …generates an individualized letter to the customer including all
4
with customer defined terms and the prompts provided by the sales rep

…prompts the AI on how to frame the message, which topics to 5 …generates meeting agenda and communication text with
include and potentially pick up on earlier interactions customer relevant discussion points

Source: Desk research; EY-Parthenon project expertise

Personalization of content to individuals’ language skills can broaden the audience by
reaching previously untapped markets, leading to higher and / or new revenue streams
16 Case example: Personalized language stylist USE CASE | ILLUSTRATIVE

Tailoring content language to user at scale Deep dive – Newspaper language personalization

Media outlets serve a diverse audience, including Especially attractive for non- One content
people with different education levels and literacy natives or for ‘translation’ of -
governmental agency text Personalized
Problem skills, making it difficult for some individuals to language
comprehend content in complex language or with
Benefits and features: Reader wants to access
technical terms ► Easier to read published newspaper article
► Summarization optionality
► Integration of dictionary
► …
Using the power of LLMs, accessibility of content can LLM
Dynamic instructions
be enhanced by personalizing its language to the
Solution A dynamic prompt selector
individual’s needs, e.g., by making it easier to read provides the language model with
using simpler vocabulary and shorter sentences Data preparation tailored instructions based on the
Breaking down long articles reader’s characteristics, providing
into smaller, digestible data context and guidance on how to
chunks for model to process personalize the article’s language
in tone and style
► Wider audience and expanded market reach from
Realized
reaching into underserved markets
business
► Improved customer experience and enhanced
impact community engagement
Content Customer data, Customer
database incl. previous content segmentation
Solutions OpenAI read, average reading time, affiliation
sophistication of text, …

Source: EY-Parthenon project expertise

Return back to meta data

Generative AI helps to easily build complex PowerBI dashboards, enabling the creation
of individual reporting dashboards in little time even for non-technical users
17 Case example: AI-tailored dashboards USE CASE | ILLUSTRATIVE

Automation of reporting dashboard creation Demo – Copilot in Power BI

Request of complex and non-standard dashboards

Problem for reporting require scarce technical resources and
are costly to develop and to maintain

GenAIs supports user to build dashboards based on

Solution chat functions, giving access to individual
dashboards for any user

► Reduced costs due reduced requirement for

Realized
specialized technical resources
business ► Better informed business decisions
impact ► Higher agility and execution speed

Solutions

Source: Desk research; Microsoft; EY-Parthenon project expertise

Return back to meta data

Introducing a GenAI-based quality assurance system can help to spot serial defects
faster and more reliably, while reducing manual effort and fostering data transparency
18 Case example: Quality assurance management

Provision of raw input data Evaluation of customer claims and Presentation of quality issues in a
as received from customers extraction of defect data via GenAI standardized dashboard
Standardized
► Scanning customer Complaints filed per month
dashboards via
claims for patterns to Customer claims forms GenAI-driven evaluation 50 PowerBI Copilot
recognize serial ► Recognizes repeating issues in claim texts
40
30
defects is a labor ► Clusters defect types
20
10
intensive and ► Attributes information to relevant databases 0
Jan Feb Mar Apr May Jun Jul Aug Sep
repetitive task
Number of flagged products: 3

► Large Language Faucet 1 7

3× ‘drips’
3× ‘sink bore size’
Automatically LLM Output
Models can 1× ‘stability‘
processed 2× ‘rust’
understand and Faucet 2 5 2×
1× ‘sink
‘waterbore size’
interpret claims forms, pressure’
Faucet 3 2 2× ‘noise‘
and distill the key Claims contents:
information from them ► Customer data Sorted and Structured Number of flagged parts: 3
► Product & parts ID clustered data reference data
Gasket ’X’ 10 Stiffens too fast
► GenAI models can be ► Free text problem description

integrated with an off- Drain ’Y’ 7 Prone to hair clogging

the-shelf dashboard to Product database Claims database Shower Links break up

5
allow engineers easy ► Product hierarchy
hose ’Z’
GenAI can attribute
and leak water

navigation when ► Historic product issues to the parts

► Product categories claims data they are likely caused by
assessing common ► Parts attribution ► Historic
claims and problems ► Product context claimant
Engineer supervision
Qualified personnel evaluates the
► Production volume data findings and either starts corrective
► Materials, etc. ► Payout data measures or discounts findings

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Improved demand forecasting accuracy has positive implications on operational
efficiency, financial planning, and customer satisfaction
19 Case example: Demand forecasting USE CASE | ILLUSTRATIVE

Demand forecasting Deep dive – Forecasting system set-up

Internal data: External data:

Obtaining accurate forecasts across different levels ► ERP / CRM system ► Weather
Input ► Store data ► Calendar
of granularity (e.g., category, region, segment) is
Problem ► Promotions ► Review sites
highly complex and requires sophisticated models ► … ► Social media
and comprehensive and reliable data ► …

Automated
data pipeline

AI-based demand forecasting recognizes interde-

Hierarchical forecasting
pendencies among quantitative factors and can Multi-level AI allows for validation of
Model
Solution further include qualitative factors from unstructured hierarchy output (top-down, middle-
data (e.g., customer reviews, social media sentiment, out and bottom-up)
etc.)

Full transparency
Output Multi-level across functions
demand forecast
► Efficient resource allocation and reduced working
Realized capital from optimized inventory management e.g., at regional e.g., at e.g., at
business ► Improved financial planning and budgeting P&L level SKU level category level
impact ► Enhanced category management
Financial Sales and
Supply Chain
planning Marketing

Source: EY-Parthenon project expertise

Return back to meta data

Product content generation (1/5)

Generative AI allows to create and customize product descriptions at scale, establishing

best-in-class product information transparency for customers
20 Case example: Product content (description and image) generation USE CASE | ILLUSTRATIVE

Product description generator Deep dive – Information content generation

Poor product information transparency leading to xxx

Problem xxx xxx xxx xxx
higher risk of product returns and low conversion rate xxx xxx
xxx xxx
xxx xxx xxx
xxx xxx xxx
xxx xxx
xxx xxx xxx

Generative AI to automate the creation of product

Product master data Existing images / visuals
descriptions at scale with content customization in
Solution
different formats (e.g. text, images, etc.) or
languages
(Pre-trained) AI model uses real
images and metadata to generate
Realized ► Enhance customer shopping experience and content in the desired style
business customer satisfaction
impact ► Increase conversion rates

Generative AI text model Auto-generated descriptive

creating info / marketing product images based on
content for online shop product information
Sales & Marketing Scaling of
IT & Technology capabilities

Best-in-class product
Solutions information transparency

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Product content generation (2/5)

Style-adaptation and translation capabilities of GPT enable generation of product

descriptions for all [Client] formats in multiple languages for each E-Shop at scale
20 Case example: Product description generation
1 2 3 4 5

Technical and style GPT models learn Product descriptions aligned Off-the-shelf Product descriptions aligned
didactics as writing styles and with format‘s communication GPT models can with format communication
database generate tailored style are generated translate text styleare generated
► LUQOM formats descriptions
differ in their target
audience and The Kaiser Idell Luxus is a
branding style in special edition of the iconic
over 17 countries Metadata lamp, which first saw the light
of day in 1931. […]

► Manually writing &

translating product

Client Client
Elegance and extravagance
descriptions into 10+ combine skillfully in the
languages incurs high Training elegant Adriana metal table
personnel costs LLM lamp. LLM

specific specific
The design contributes to the Sample AI
charm of the high-quality translated
► GPT models can lamp. The elegant lamp with a description
stable base and a wide
replace the manual lampshade conjures up a
[...] Eleganz und Extravaganz
vereinen sich gekonnt in der
creation and bright light that makes it easy eleganten Adriana Metall
translation of to read & work. Tischlampe. Das Design […] trägt
zum Charme der hochwertigen
product descriptions Lampe bei. [...] die elegante Lampe
Product A classic notary lamp with a mit stabilem Fuß und breitem
Lampenschirm zaubert helles Licht
descriptions black finish and mirror glass. hervor, das das Lesen und Arbeiten
Wow, what an eye-catcher! …. erleichtert.

Automated content generation at scale reduces marketing and overhead costs

Source: Desk research; EY-Parthenon project expertise

Product content generation (3/5)

Generative AI enables the creation of product images on a large scale, allowing

personalized experience for customers and facilitating the maintenance of the webshops
20 Case example: Product image generation USE CASE | ILLUSTRATIVE

Deep dive: Webshop imagery generation & individualization

“Contemporary style”

Pre-set styles
► Manually editing all
product images is not Country house style (Pre-trained)
GenAI uses Customer 1
feasible, so some products
are shown without fitting
context
Industrial style default product
image, available Client or

specific
Minimalist style metadata and “Cozy farm style”
user prompts to
► Customers have
difficulties relating to
products shown in front of
Client
Default image on
Client
Scandinavian style generate images
in the desired
a neutral background
specific
product landing
page specific … style
Customer 2

► Allowing customers to
generate previews of Custom styles
or
products in different
environments increases
likelihood of purchase “
Place this lamp in an
upscale French
Customer X …

country house
bedroom with beige- Not required if
white walls, next to a preset is chosen
Customers receive individual
wooden nightstand.”
product presentations that fit
their ideas for the room

Source: Desk research; EY-Parthenon project expertise

Product content generation (4/5)

GenAI can be leveraged to build product descriptions automatically, optimize them for
SEO and eventually adapt them specifically to individual customer profiles
20 Case example: Product description (incl. SEO optimization) generation USE CASE | ILLUSTRATIVE

Cumulative GenAI product description writing use cases

A B C

Product description generation SEO-optimized product listing writer Customer-specific product description

Automated product description to Advanced recommendation based on Persona based, dynamic product presen-
improve information for purchase decision incoming customer feed and SEO tation with pictures and text description

Lorem ipsum dolor sit amet, Etiam vitae tempus lectus.

consectetur adipiscing elit. Proin Donec mattis, felis eu aliquam
sollicitudin lectus risus, id tempus, ipsum ligula porttitor
sagittis massa lobortis sed. nulla, quis gravida leo enim eu
Nulla nec est erat. Sed non diam mi.
ac nisi venenatis faucibus.
Mauris ex nulla, pretium nec
Nullam pharetra nisl leo, laoreet
tincidunt eu, rhoncus nec purus.
dapibus velit diam aliquet odio.
Based on customer specific characteristic,
e.g. age, level of education etc. product
descriptions will be personalized, e.g.
short vs. long text, language, writing style
etc.

D Monetization opportunity: Offer GenAI-enabled product description as a service to platform partners

Source: EY-Parthenon project expertise
Product content generation (5/5)

EY’s Generative AI Product Listing Generator automatically creates new content across
large portfolios in a fraction of the time of the manual process
20 Case example: Product description (incl. SEO optimization) generation USE CASE | ILLUSTRATIVE

 Represents the voice of the customer: Builds product descriptions that represent the voice of the customer in a dynamic way
Value
 Reduced time for content creation: Creates content across large portfolios in a fraction of the time of the manual process
drivers
 Improved keyword generation: Introduces keywords in the product reviews that help with search engine optimization

Source: EY-Parthenon
Factory location network optimization (1/1)

AI driven network optimization mitigates the risks arousing from volatile market
environment by considering 360-degree supply chain factors
21 Case example: Factory location network optimization USE CASE | ILLUSTRATIVE

Optimization of the factory location network Deep dive – AI driven network optimization
In-depth understanding of
underlying business logic Multi-layer solution
Customers required
Products / Components
► Demand ► Which product
Volatile market environment driven by increasing ► Proximity to markets assembly to be
OM
uncertainties across various strategic dimensions, ► … shifted?
Problem ► Which components are
incl. geopolitical or global trade, force manufacturing
required?
companies to review their factory network set-up Facility capabilities Optimization model to
► Production capacities find best possible
► Produced product lines solution out of feasible Facility capability shift /
► Production costs
option space expansion
► … ► Where can the
production capabilities
Local infrastructure be utilized?
Apply AI to simulate various supply chain disruption ► Transportation networks
NM ► Is there any expansion
required?
scenarios and estimate their probabilities, deriving ► Access to distribution
Solution hubs
ideal factory locations with optimal degrees of ► Energy supply and costs Network model for
Net financial impact
manufacturing resilience and flexibility ► … multi-knapsack and
► What is the required
salesman problem
Macroeconomic factors investment and
optimization associated returns
► GDP growth
► Inflation looks like?
► Interest and exchange ► What are the

rates ML opportunity costs?

► …
► Increased resilience in supply chains and
Realized Net risk impact
operations Geopolitical risks
business ► Global trade Machine learning ► Which risks are
► Enhanced flexibility in production across ► Policies model to perform mitigated?
impact factories ► Conflicts or sanctions complex market ► What are the internal and

► … Individual modelling tasks external effects of the

workstreams changes in network?

Source: EY-Parthenon project expertise

Return back to meta data

Inventory & procurement optimization (1/1)

Optimize working capital while ensuring production capability through AI-based

evaluation of macroeconomic factors, enabling cost minimization and revenue growth
22 Case example: Inventory & procurement optimization USE CASE | ILLUSTRATIVE

Minimize working capital while ensuring production ability Deep dive – AI-driven inventory optimization

Production plan data: Supplier contract data: External data:

► Production cycles ► Contractual details ► Geopolitical activities
External factors lead to supply chain disruption and ► Needed amount per input ► Purchase quantities ► Interest rates
volatility of availability. Changing influencing fac- Input factor ► Prices ► Governmental regulations
Problem ► Available inventory
tors and its (combined) effects complicate the deter- ► … ► Social media
► Cost of production ► Supplier reliability
mination of an optimal inventory level per product downtime ► …
► …

Automated Automated Automated

data pipeline data pipeline data pipeline

Use AI to calculate the optimal purchase quantity and Calculate Calculate Calculate
AI score per AI score per AI score per
Solution intervals for each product, taking into account all Model
product product product
relevant factors and their developments

AI Evaluation with inclusion and combination

of all relevant input factors

► Cost minimization through working capital

Realized
optimization Procurement Procurement Shopping carts
business ► Potential revenue growth through closing of quantity cycle (product
impact production gaps bundles)
Output Optimized inventory
automated supply process

Source: EY-Parthenon project expertise

Return back to meta data

Regional success prediction (1/1)

AI can facilitate the automation of regional success prediction when planning the
deployment of new technologies, offering a prioritization and concise recommendations
23 Case example: Large-scale regional success prediction USE CASE | ILLUSTRATIVE

Regional success prediction Deep dive – Regional success prediction at scale

Interactive dashboard
User-friendly access to prediction results
Roll-out of new technologies like mobile 5G- or fiber- with the ability to further interact with
infrastructure is a complex and expensive process, information
Problem
requiring an efficient management and allocation of
available resources AI-powered prediction model
Evaluation of the financial potential and risks associated with the
decision expand or not to expand in both types of locations with strong
market position (‘home turf’) and weak positioning (‘competitor
territory’) based on region-specific characteristics and revenue drivers

Automated regional success prediction for e.g.,

technology roll-out can provide clear recommend-
Solution What are the What are customer How is the company
dations and prioritizations through an interactive How is the competition
population preferences? How is positioned in the
dashboard displaying all relevant information positioned? What do
characteristics in the churn rate impacted region? How did they
they offer?
region? in each scenario? perform in the past?

Input factors

Realized ► Enhanced decision making based on data-driven

Local population Customers Capabilities Competition …
business selection of locations ► Population size ► Preferences ► Existing ► Market shares
impact ► Enablement of continued and sustainable growth ► Demographics ► Cons. behavior infrastructure ► Existing
► Purchasing power ► Churn rates ► Past performance infrastructure
► … ► … ► … ► Service offerings
► …
Source: EY-Parthenon project expertise

Return back to meta data

Internal process efficiency (1/2)

We can support by providing a profound understanding of business processes and AI

applications to identify the right investment and implementation areas for automation
24 Case example: Diagnostics – AI for process efficiency USE CASE | ILLUSTRATIVE

AI for process efficiency Example – Efficiency Potential in Demand to Cash Process

Efficiency Potential with GenAI

Research &
Pharma companies have in-depth domain knowledge, Demand to Cash <10% 10-20% >20%
Development Low Medium High
but often require external expertise in the areas of
Problem Demand planning
technology and data to identify less apparent use
cases of AI to enhance process efficiencies Supply planning
Regulatory
Sales order management

Supply Chain
Client specific Select and manage logistics service
providers
Following a structured approach, business processes (behind the placeholder there is a content relevant for Pharma/ Life Science sector and
Demand to Cash Process. Ideas for process efficiencies are presented on the next slide)
are analyzed and evaluated regarding their potential Plan transportation
Solution
for increased efficiency through the implementation
Billing
of AI applications Manufacturing &
Operations Manage and process collections

Manage cash application

Monetization &
Customer care management
► Enhancing in process efficiencies across all Commercialization
Realized
processes and functions
business ► Improved resource allocation and investments
Total efficiency potential in Demand to Cash
impact ► Increase in process security and reliability Monitoring & process from AI applications and
Reporting automation at ~20% and higher

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Internal process efficiency (2/2)

We identified a large number of processes with efficiency potential of up to 20% and

more leveraging the power of Generative AI
24 Case example: Internal process efficiency USE CASE | ILLUSTRATIVE

Overview Process Efficiencies Efficiency Potential with GenAI

<10% 10-20% >20%
Low Medium High

Cost saving potential of >20%

in all typical corporate
processes

Client specific
Customer landscape analysis (1/1)

Construction distribution landscape has a blackhole in terms of transparency on

potential B2B customers, which can be systematically improved with the help of AI
25 Case example: Customer landscape analysis USE CASE | ILLUSTRATIVE

Landscape analysis on craftsmen businesses Deep dive – transparency of B2B customers

High number of small craftsmen firms and Manufacturers Distributors Processors Customers
increasingly blurred expertise makes it difficult to
► Wholesale
obtain a holistic player overview for a structured Direct ► Service providers ► Industrial
Problem sales approach
► Specialized ► Installers clients / project
Whole- Manufacturer wholesale
customers
sale ► Online
► Private
► Intermediaries Highest level of
Retail customers
transparency

Transparent (potential) provider and customer landscape

Solution
Client /Sector specific
A customer intelligence solution with AI can provide
a detailed view into the market landscape and
Data aggregation / transformation

(behind
customer potentials by providing the placeholder
transparency Systematically
on there is a content consolidate
relevant for datamarket)
Construction sources with advanced AI (language) models such as web
crawling, web ads, handworker association databases, governmental source (e.g.
existing companies, their size, performance, etc. Bundesanzeiger) to collect installer and provider characteristics to obtain transparency

 Increased sales from optimized sales approach

and easier identification of growth potential Tools &
HVAC Plumbing Electricals
Realized  Comprehensive understanding of player landscape
supplies

business and enhanced customer transparency Inst.

Ventilatio Ground& Installatio Wires& Lightin
impact Heating
n
AC
sewage
Sanitary
n
materia
cables g
… …
ls

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Consumer insights (1/3)

GenAI can create consumer insights and quickly identify trends through its global
language competencies and social listening
26 Case example: Consumer insights USE CASE | ILLUSTRATIVE

Consumer insights Deep dive – understanding consumers

CPG companies lack direct contact to consumers due Global web Product reviews
to POS sales making it hard to understand and crawling of and related
Problem
identify consumer trends, preferences, or the (social) media comments
optimized product portfolio allocation

Use AI big data capabilities to collect consumer data,

Solution analyze consumer behavior, identify buying
preferences, and forecast global shopper trends
Global data is processed through global language
competencies of AI, analysis of social buzz conducted, and
respective trends in countries / regions and related
► Earlyidentification of relevant trends consumer preferences are identified
Realized
► More effective marketing and promotion
business campaigns
impact ► Improved and consumer suiting product portfolio
Based on consumer preferences and
evaluated reviews, preferred products
and brands are identified
Solutions

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Consumer insights (2/3)

AI-driven algorithms empower [Client] to make data-backed presentation decisions by

providing early, actionable insights into consumer demand
26 Case example: Consumer insights USE CASE | ILLUSTRATIVE

Consumer insights Deep dive – understanding consumer behavior

Consumer insights Consumer profiles

Consumer trends, preferences and
Global web Product reviews
Problem segmentation are highly dynamic and require crawling of (social) and related
large amount of resources media comments

Buyer information Buyer behavior Transactional data

Use AI capabilities to collect consumer data, (demographics, etc.) (search activity) (purchases, etc.)
analyze consumer behavior, identify buying
Solution Data input through
preferences, and forecast global shopper retailer information
trends based on real-time global data
Global data is processed through global
Historically collected buyer information
► Earlyidentification of relevant trends language competencies of AI, analysis of
and transactional data is processed
Realized ► More effective marketing and promotion
social buzz conducted, and respective
trends in countries / regions and related
business campaigns consumer preferences are identified
impact ► Improved and consumer suiting product
Creation of consumer profiles / personas
presentation  Identification of consumer segments
Based on consumer preferences and (e.g., demographics etc.)
evaluated reviews, preferred products and  Recognition of suitable products for each segment
Impacted Sales & Self- brands are identified  Filled out standardized consumer templates
business Key developed
functions Account growth
Globally and locally optimized Identification of trends suitable for each
product presentations consumer (segment) and optimized consumer
Solutions targeting

Source: EY-Parthenon
Consumer insights (3/3)

Today‘s language models are fully capable of designing comprehensive and insightful
consumer surveys, allowing for automated and personalized questionnaire design
26 Case example: Consumer insights (survey) USE CASE | ILLUSTRATIVE
“Commercial performance of the [Henkel Syoss] brand has been mixed recently and new consumer insights are required to boost performance.
Please design a concise but insightful and actionable consumer survey to learn about levers to improve commercial performance.
Make sure that competitive and value proposition are reflected on top of product improvement considerations.”

Brand / Topic specifics

(behind the placeholder there is a content relevant for Henkel Syoss brand)
Urban digital twin (1/1)

Smart city planning help local regulating bodies to get more out of their assets and
create room for partnerships with private sector
27 Case example: Urban digital twin USE CASE | ILLUSTRATIVE

Urban digital twin Deep dive – Smart city planning

Increasing population in urban areas and tightening Safety Infrastructure

supply around central city utilities create bottle ► Traffic safety ► Road and mobility network expansion …
Problem ► Optimized emergency response ► Healthcare access points
necks, exhaust the city infrastructure and affect
quality of life
Mobility Cost of living
► Public transport load balancing ► Land usage and permitting …
Artificial intelligence to aggregate complex data and ► Intelligent traffic management ► Dynamic utilities pricing and tracking

Solution model ‘digital twin’ for a city deriving scenarios for

the infrastructure investment and planning to meet
future load AI AI AI AI AI

► Integrated data solutions enabling urban Models AI driven 3D digital twin: Identifies dynamics, dependencies and recurring
Realized
expansion inhabitant centric services patterns within a city to improve multiple aspects of urban quality of life
business
► Urban development in environment and cost-
impact efficient way
Data Data retrieval and integration
retrieval Aggregating and normalizing data suitable for the 3D representation

Strategy & Innovation

Disruptive
Sales & Marketing in Geospatial Infrastructure Building layouts Vehicle sensors Transportation
growth Data
private sector acquisition
networks

► Aerial surveys ► Public records

► IoT data ► International databases

Source: EY-Parthenon project expertise

Return back to meta data

Construction multi-dimensional modelling (1/1)

Construction industry is yet to reach data maturity levels to leverage “Building

Information Modelling” to its fullest, improving construction process for all stakeholders
28 Case example: Construction multi-dimensional modelling USE CASE | ILLUSTRATIVE

Multi-dimensional modelling Deep dive – Building information modelling

6D-BIM 7D-BIM
Industry experiences increased demand for 3D-BIM 4D-BIM 5D-BIM
Facility management
Quantities & Costs Sustainability
Problem sophisticated data to build digital twins and 3D / 4D / 3D-modelling Time / Schedule

5D BIM construction simulations ► What are the ► What should be ► What are cost ► What are ► What are
design the sequence of estimates and suitable required
requirements construction implications per solutions that maintenance
Utilize AI to improve the quality and availability of and building activities? stakeholder? yield a low steps?
data by extracting and consolidating product characteristics? ► Where are time- ► What are the energy ► What are the
Solution ► How to minimize dependencies in associated consumption? expected
information from manufacturer's data sheets or other planning errors? the construction risks? ► How to best maintenance
reliable data sources process? guarantee costs over the
quality, life cycle?
comfort, and
sustainability?
► Improved data ecosystem with cross-effects on Model
Realized AI model aggregates and defines the relationship among data on
other areas, incl. enhanced sales, improved AI
business product descriptions, etc.
various dimensions and granularity and identify dependencies,
providing foundation for advanced scenario analysis in BIM
impact ► Competitive advantage from superior data quality

Architectural Mechanical Costs Sustainability Structural

What is the floor What are the What are material How high is the What are the
Sales & Defensive plan? installation plans? and overhead energy maintenance
Marketing growth costs? consumption? regulations?

Product data sheets Supplier data

Construction plan …
Solutions OpenAI

Source: EY-Parthenon project expertise

Return back to meta data

Automated tender evaluation (1/1)

Generative AI can be used to increase bid success chance and reduce processing costs by
automizing the bid identification and vetting process
29 Case example: Automated tender evaluation USE CASE | ILLUSTRATIVE
Provision of input and Evaluation of available tenders Ranking of opportunities by expected
reference data using generative AI value, product fit and relationship
(incl. provision of context details)
► Manually scanning
tender databases Tender opportunity database
Generative AI interprets tender
incurs high personnel
input data and extracts key
costs and relies on information for analysis and
each salesperson’s Tender ranking overview
ranking
subjective judgement Raw input Weighted Est. Product Prev. co-ops
Tender name ...
value fit with planner
ranking logic
Schmidgasse 4 €2.0m Strong 3 cooperations ...
► Generative LLMs
Müllerstr. 12 €0.5m Strong 5 cooperations ...
combined with
traditional AI models LLM Grünallee 30 €2.0m Low 1 cooperation ...

can automate that Bergmannstr. 7 €0.8m Medium 0 cooperations ...

process along pre- Kaiserplatz 16B €1.3m Low 1 cooperation ...
Public information (internet)
defined criteria Reading aid:
► Reputation of project / developer Context
“Schmidgasse 4 has the highest EV among
► Automation of tender ► Roadblocks (e.g., lawsuits) Evaluation parameters projects with strong product fit and for which we
identification ► Estimated contract value1 know the planner, so it should be prioritized”
increases chances of Historic project documentation ► Expertise and product fit
bid success (by finding ► Project general data Historical ► Previous exp. with
data developer
better fitting tenders) ► Contract size, scope and type
and increases revenue ► Developer relationships ► Future follow-up potential
potential ► Challenges ► Project prestige

Automated tender evaluation at scale increases revenue potential while reducing costs for manual analysis
1.Estimated contract value = Overall project value * % servable by own portfolio
Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Virtual shopping guide (1/1)

A virtual product guide supports customers in finding the right product for their needs
and provides opportunities for personalized cross- and up-selling while customer shops
30 Case example: Virtual shopping guide USE CASE | ILLUSTRATIVE

Virtual shopping guide Deep dive – System set-up

Responses can include:

► Product suggestion
Due to store size and differing layouts, customers
► In-stock information
have difficulties finding the right products quickly ► Location & navigation
Problem Customer asks question in
when shopping in the store, leading to increased need ► Adjacent products
his own words to chatbot
for personal support by staff

Additional instructions
LLM
Introduction of additional logic
for product selection, e.g.,
preference for private label,
Virtual product and store guide that identifies the high margin products, etc.
Solution suitable product(s) for customers’ purposes and Data retriever
navigates them to it in the respective stores

What is the right Is the product Where is the Which other What are the
product for the currently on stock product located in products did customer’s
customer’s needs? in the store? the store? people buy? preferences?

► Enhanced shopping experience and satisfaction

Realized ► Reduction of efforts and capacity required for in- PIM, product
business store customer support master data, Inventory
Store layout and Basket analysis, Customer data,
merchandise
impact ► Increased opportunities for personalized cross- product manuals stock data recommender historic purchases
system1 system
and up-selling Refer to data
quality case
Information retrieved from various data bases
1.Raster plan
Source: EY-Parthenon project expertise

Return back to meta data

AI-based return and accrual prediction (1/1)

Numerical AI helps to precisely predict required financial accruals for return handling in
e-commerce while reducing operational effort by automated creation of reports
31 Case example: AI-based return and accrual prediction USE CASE | ILLUSTRATIVE

Prediction and reporting of returns accruals Deep dive – Process flow

Purchase
High fluctuation in online retail return rates due to Consumer triggers order in online shop
Problem seasonality and product group, resulting in high
provisioning shortfalls

Artificial intelligence
Calculation of individual probability of product returns
based on product attributes and consumer behavior

Leverage AI to calculate an individual return

Individual
Solution probability per order and automatically generate the return
accruals and controlling reports probability

Artificial intelligence
Aggregation of return probabilities across all orders
and creation of returns accruals per brand / division
as well as generation of financial reports
Realized
► Improves working capital by managing provisions
business Improved provisions
Financial figures
► Increases process security and reliability
impact and reports
management with direct
effects on working capital Precise proposal for
financial accruals
Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Assortment planning optimization (1/1)

Deploying AI will improve the alignment of the product range with customer preferences
while accounting for restrictions, boosting sales and cutting excess warehousing costs
32 Case example: Assortment planning optimization USE CASE | ILLUSTRATIVE

Assortment planning optimization Deep dive – Assortment planning optimization

Selection of products for introduction and phase-out

Problem across different levels of granularity (e.g., category, Outcomes can include: Procurement team evaluates
region, segment) requires high manual effort ► SKU inventory increase / decrease current product range
► SKU phase-out
► New product introduction

AI

Leverage AI to optimize inventory levels and phase-

outs in existing product portfolio and integrate
Solution Data retriever
external data (e.g. through web crawling and macro
trend analysis) to identify new products
What are the
What is the
Which legal shelf area What are
What were assortment What are the
restrictions on restrictions substitute and
historic rotation width and depth current market
SKU stocking per SKU complementary
rates per SKU? in each trends?
exist? category per products?
category?
Realized ► Enhanced category management store?

business ► Improved financial planning and budgeting

impact ► Improved customer satisfaction
Contract
Product data Web crawling of
Sales data Product data Store data (descriptions and (social) media,
information
images), supplier product reviews
info
Source: EY-Parthenon project expertise

Return back to meta data

In-store execution (1/1)

With the support of AI image analysis, in-store execution can be monitored more
efficiently and frequently
33 Case example: In-store execution USE CASE | ILLUSTRATIVE

In-store execution Deep dive – Perfect store execution

Monitor and gain … with crowdsourcing … with shelf cameras /

In-store execution audits with own field teams insights on … through consumers robots operated by retailer
Problem represent costly and highly manual efforts ► Assortment
resulting in lack of control / contract enforcement ► Shelf availability
► Pricing
► Shelf
planogram
► Promotion placement
► Competitor positioning
Apply AI image analysis to compare in-store images ►…
that are taken by consumers, shelf cameras or
Solution
robots (operated and data sold by retailer) to gain
insights on store condition (frequently and at scale)
Contract data
AI image analysis
analysis
AI Analysis of in-store
Pricing, planogram,
execution and
promotions,
Realized ► Saved costs through reduction of field teams conditions
assortment, etc.
business ► Continuous and real-time insights on in-store
impact execution

First hand insights into store conditions and clarity on execution of contract
Solutions agreements through AI image analysis

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

New business models from car sensor data (1/1)

Available car sensor data hold high potential for new business models – Utilizing AI and
deep learning techniques can help to realize value and establish new revenue streams
34 Case example: New business models from car sensor data USE CASE | ILLUSTRATIVE

Using car sensor data to identify new business models Value realization – Rental car ‘Driver score’

Car sensor data

► Speed
Different customers have different demand profiles,
► Acceleration Real-time data transfer requires
requiring tailor-made solutions and pricing models
Problem ► Breaking connection to internet from
for the products and services they consume (e.g.,
► Steering behavior telematics devices (in-vehicle,
‘pay as you go’ or ‘pay how you drive’) ► Time of day plug-in device, or mobile app)
► Location (GPS)
► Distance driven

AI scoring model
Use AI and advanced deep learning techniques to AI Evaluation of individual’s driving behavior to
identify patterns and trends in the data, serving as identify how drivers have ‘treated’ the rental
Solution car
foundation for creation of new business models that
meet individual customer demands at scale
‘Driver score’ as reference for damage claim resolution
► Addressing customer concerns about unfair treatment
regarding damage claims, rental car companies can apply a
‘Driver score’
► New revenue streams from business model ► Score enhances transparency and reduces perception of being
Realized expansion exploited by rental car company
business ► Improved customer experience from better ► Serves as reference point when evaluating damage claims,

insights into product use and customer allowing for more equitable resolution process
impact
preferences Subscription model interesting for rental car companies or
insurances companies (usage-based insurance)

Source: EY-Parthenon project expertise

Return back to meta data

Customized discounts and product promotions (1/1)

AI-aided customized discounts and promotions system can promote sales by increasing
customer conversion and retention rates while increasing marketing efficiency
35 Case example: Customized discounts and product promotions USE CASE | ILLUSTRATIVE

Customized discounts and product promotions system Deep dive – Real-time shopping promotions

► Demographicinformation
Intense competition and high price transparency in
► Transactional
data
Problem ecommerce lead to increased rates of abandoned
► Engagement data
baskets and customer churn Customer ► Behavioral data Basket
data ►… value
AI model determining
Neural network predicting
promotion option
Customized discounts and promotions for eligible probability of basket abandonment
range for basket
Solution shoppers / customers that show risk of basket
composition
abandonment / churn but high value / LTV Probability for basket Feasible promotion
abandonment options (e,g. %-
discount, bundles etc.)
Realized ► Increasedcustomer conversion and retention rates Re-calculation of basket abandonment
business ► Improved promotional budget allocation and probabilities for high risk customers with
impact marketing efficiency high basket value under feasible options
Shopper can be informed
about discounts and Selection of optimal promotion
promotions through e.g., option according to pre-defined
Operational pop-up windows or decision rules (e.g., cost-benefit)
Sales & Marketing basket reminder email
efficiency
Customized discount &
promotions system per shopper
► Optimized allocation of promotional budget
► Increased conversion to buying customers

Source: EY-Parthenon project expertise

Return back to meta data

Deal sourcing and diligence (1/1)

PE firms can realize major efficiencies during deal sourcing and diligence by deploying
GenAI, fully or partly automating most of the labor-intensive process steps
36 Case example: Deal sourcing and diligence USE CASE | ILLUSTRATIVE

GenAI in deal sourcing and diligence Deep dive – Application potential of GenAI
GenAI potential GenAI action
Today’s deal sourcing and diligence process is Data sourcing from e.g.,
Research for deal sourcing
Problem structurally repetitive and involves significant CapitalIQ
Initial screening GenAI Classification for eligible target
amounts of manual labor
Preliminary due diligence Processing of public information
Preliminary analysis Performing pre-defined analyses
Leverage GenAI by integrating the technology into
the process, thereby fully or partly automating Initial DD GenAI Drafting of reports
Solution If required, requests additional Auto-generates
individual steps, incl. deal origination, analyses information requests Drafting of emails
With PE
conduction, risk assessment, and target valuation Financial, commercial, legal, and ops DD oversight Drafting of reports

Major DD efficiencies: Final risk assessment based on DD GenAI Performing risk analyses
Realized ► Increasing target insights and market Target valuation GenAI Modelling based on business plan
business assessments
Investment negotiation PE
impact ► Decreasing time between sourcing and deal close
If consensus,
► Potential for lower deal-related expenses GenAI Drafting with limited fine-tuning
preparation of legal docs and obtaining approvals
Investment completion PE

Strategy & Scaling of Once approved,

PE
Innovation capabilities negotiation of deal structure with target
Preparation of legal docs Drafting with limited fine-tuning
GenAI
Fulfilment of regulatory requirements Drafting with limited fine-tuning
Deal closing PE

Source: EY-Parthenon project expertise

Return back to meta data

Deal sourcing and diligence (2/2)

Investing in AI driven solutions in pre-deal project work is crucial to stay on par with
competition and improve lead generation as well as the final deliverable for our clients
36 Case example: Deal sourcing and diligence USE CASE | ILLUSTRATIVE

Pre-deal Insight paper Pre-deal Due Diligence

► With AI leverage previous project documents about a potential ► Main deliverable of DDs are highly standardized and could be
target (targets usually remain the same and we probably did a generated more efficiently
XDD), source financials from CIQ, market reports from
databases ► Goal is to leverage existing knowledge by incorporating
previous project work and reduce manual repetitive research
► Goal is to reduce resource allocation to 1 week to tweak AI to use resources for more cognitively value-adding tasks
generated deck (compared to 3-4 week atm)

Common repetitive modules require only

one-time investment to improve efficiency

1 2 3 4
Defining target lists and Retrieving information from Content aggregation and Generation of standardized deck content
information sources internal and external sources tuning of output ► Company overview
► Asset list of PE / VC Current CBAT tool is an ► Business model / area deep-dives
clients initiation for automizing this ► Perimeter of products and services
► Capital IQ step ► Market size and forecasts
LLM
► Factiva ► Financial overview
► EY Embryonic ► Latest developments in the market
► Mergermarket
► Previous project Could be used to identify potential
documents risks for the assets in holding period
Content (see PE Fund case)
► Previous Insight papers
databases
Drug demand forecasting (1/1)

Leveraging AI for demand forecasting leads to improved accuracy and data-driven

decision making for operational and financial planning, as well as sales and marketing
37 Case example: Drug demand forecasting USE CASE | ILLUSTRATIVE

Demand forecasting Deep dive – System set-up for demand volume forecasting

Company
► Sales volume
Forecasting of drug demand is currently performed
► Cannibalization rate (if new
with a top-down and bottom-up statistical approach, Full transparency
product introduced)
Problem across functions
requiring a lot of manual work and relying largely on ► Promotions and advertisement
Financial planning
the individual’s experience, rather than on facts
Disease ► What are the required

► Prevalence rate of disease

investments?
► How does it affect working
► Symptom rate
AI capital requirements?
► Diagnosis rate

Apply an AI-based demand prediction model to Supply Chain & Operations

Drug ► How to plan available
generate accurate forecasts across various levels of Neural network
Solution ► Treatment rate production capacities?
granularity based on a comprehensive and reliable set ► Drug accessibility to perform multi-level ► Are existing suppliers
of data volume forecasting, aligned with forecast?
identifying deeply
Competitors
embedded patterns
► Market share of drugs Sales & Marketing
and dependencies
► Number of competitors / drugs ► Which products to push at
while allowing doctors?
► New launches and latest
validation of output ► On which drugs to allocate
innovations
► Better and faster decision-making (top-down, middle-
Realized Society
marketing resources?
► Improved financial and operative planning out, bottom-up)
business ► Population size
► Efficient resource allocation and reduced working
impact capital from optimized stock management ► Adherence / compliance with
subscription
► Persistence with drug therapy

Source: EY-Parthenon project expertise

Return back to meta data

Drug Discovery (1/5)

Benevolent PlatformTM is an ML-enabled and hypothesis-driven platform to accelerate

drug by improving disease understanding and target identification
38 Case example: Drug Discovery – Disease characterization Target validation USE CASE | ILLUSTRATIVE
Disease Target Target Hit Hit-to-lead Preclinical
Activities 1 characterization 2 2
identification 3 3
validation 4 generation4 5
optimization1
6 testing

Description (Benevolent Platform ) Data input2 and output

► 50%
clinical failures due to poor understanding of Data input (exemplary)
Challenge / disease biology / mechanisms
Clinical data: Clinical and omics-
pain point ► Target discovery often not scalable or repeatable ► Assay data databases:
► Disease-specific approaches are prone to bias ► Trial data ► Diseases
• AI-driven NLP and ML (Knowledge Graph approach) to: Literature: ► Symptoms
• Perform inference and reasoning analyses on large ► Journal articles ► Genomics
biomedical datasets to enhance disease ► Patents ► Proteinomics
AI solution
understanding
• Mine data and discover high-conviction targets Real-time in silico experiments via NLP and ML
(disease and modality-agnostic)

• >2x probability of success in drug development

Business
• ~2 years shorter target-to-candidate timeline
impact ML ML
• ~$18m savings for target-to-IND development

Current user(s) AI provider

User and
provider Detailed Novel target Selected most-
disease map candidates promising targets
1) also includes “Lead selection”; 2) AI-harvested public domain and Benevolent AI’s own experimental data, both preclinical / clinical and in silico
Source: company website; company reports; EY-Parthenon analysis; Abbreviations: IND = Investigational new drug

Return back to meta data

Drug Discovery (2/5)

PandaOmics and Chemistry42 are generative AI-driven platforms designed to identify

targets and generate leads with higher probability of success rapidly and inexpensively
38 Case example: Drug Discovery – Target identification Hit-to-lead optimization USE CASE | ILLUSTRATIVE

Disease Target Target Hit Hit-to-lead Preclinical

Activities 1
characterization
2
identification
3
validation
4
generation
5
optimization1
6
testing

Description (PandaOmics; Chemistry42) Data input2 and output

► Target discovery often not scalable or repeatable and Data input (exemplary)
Challenge / often focuses on single targets Literature: Molecular Omics-databases:
pain point ► Drug discovery projects still require 1000s of molecules ► Journal articles databases: ► Genomics
be made / tested over many years before clinical trials ► Patents / Grants ► Molecules / ► Proteomics
► Trial data Biologics ► Epigenomics
► AI-driven suite to identify / validate novel targets, then
► Transcriptomics
design / optimize molecules for the targets before
preclinical testing
AI solution
► Hit generation and optimization is fully automated and
lead compounds are ranked across various pre-clinical Generative AI-driven identification, validation, design and optimization
success metrics (e.g. potency, stability, safety)

• Hit generation and optimization in 1 week vs. 2-4 years

Business traditionally
impact • Discovery to preclinical candidate in 18 months for
~$3M vs. 3-6 years and $600-800M traditionally

Current user(s) AI provider

User and
provider

1) also includes “Lead selection”; 2) data sources are both publicly available and proprietary
Source: company website; company reports; EY-Parthenon analysis
Drug Discovery (3/5)

Chroma explores patterns in protein structures and generates novel protein molecules,
adhering to pre-specified constraints
38 Case example: Drug Discovery – Hit generation USE CASE | ILLUSTRATIVE

Disease Target Target Hit Hit-to-lead Preclinical

Activities 1
characterization
2
identification
3
validation
4
generation
5
optimization1
6
testing

Description (Chroma) Data input2 and output

Generating protein-based drug hits is significantly more Data input (exemplary)
Challenge / complicated vs. small molecules due to 1) increased Protein database:
pain point structural complexity; 2) bioavailability / stability ~200k high-resolution
challenges
public protein
structures and next-
• Described as “DALL-E 2 of biology” - utilizes diffusion gen sequencing with
modelling (via text-to-image AI) to generate large, ~200m amino acids
AI solution novel protein libraries – based on publicly available
Protein Data Bank – that can be screened against
Learning from 3D protein structures and amino acid sequences
potential targets
Generation Design
• Can generate 10k+ protein variants and test each Collapsed Refined Complete
system system system
through cell-based assay vs. 100s at current industry Training
Business standard
impact
• Increased speed (due to self-improving ML / automation)
• Reduced resources (due to open source data input)
Current user(s) AI provider
User and
provider Not reported Novel proteins generated
1) also includes “Lead selection”; 2) data source is external and publicly available
Source: company website; company reports; EY-Parthenon analysis
Drug Discovery (4/5)

MatchMaker and POEM utilize ML to predict extensive small molecule-target

interactions and support ADMET model generation
38 Case example: Drug Discovery – Hit generation  Hit-to-lead optimization USE CASE | ILLUSTRATIVE
Disease Target Target Hit Hit-to-lead Preclinical
Activities 1 characterization 2 2
identification 3 3
validation 4 generation4 5
optimization1
6 testing

Description (MatchMaker ; POEM ) Data input and output

► Typical in silico discovery only focuses on single target, Data input (exemplary)
Challenge / leaving plethora of unexplored potential Internal databases: External data2:
pain point ► Available predictive tools lack standardization of
► Polypharm Database ► Drug databanks
ADMET properties, limiting reproducibility
► Proteome and NGS

• AI tool to match structures of millions of compounds to ► Journal articles

150k proteins for extensive hit generation /
AI solution optimization
• The approach includes ADMET prediction for outlining Real-time data analysis with ML
relationship between chemical structure and activity

• Achieved over 60% actionable hits for pharma clients –

Business
significantly outperforming traditional quantitative ML
impact
structure-activity relationship approaches

Current user(s) AI provider

User and
provider MatchMaker - predicted small POEM - predicted ADMET
molecule-target interactions proprieties of small molecules
1) also includes “Lead selection”; 2) external data sources are both publicly available and proprietary
Source: company website; company reports; EY-Parthenon analysis; Abbreviations: ADMET = Absorption, distribution, metabolism, excretion and toxicity; NGS = Next generation sequencing
Drug Discovery (5/5)

Bio-AI platform is coupled with "Patients-on-a-chip" to reduce the need of animal

testing and better predict clinical safety of thousands of potential molecules
38 Case example: Drug Discovery – Preclinical testing USE CASE | ILLUSTRATIVE
Disease Target Target Hit Hit-to-lead Preclinical
Activities 1 characterization 2 2
identification 3 3
validation 4 generation4 5
optimization1
6 testing

Description (Bio-AI clinical prediction platform) Data input2 and output

Data input (exemplary)
Predicting drug safety (esp. for diseases that with no
Challenge / Patients-on-a-chip: External databases:
animal models) is challenging, delaying clinical trials and
pain point ►Data collected by
costs companies billions of dollars ►Databases and literature
nano-sensors on on compound safety and
miniaturized organ- efficacy
• ML to better predict clinical safety and efficacy by
system replicas
testing known drugs / candidates on “Patient-on-a-chip”3
AI solution • AI uses results to re-train and differentiate safe
compounds from unsafe compounds Real-time data analysis with ML
• AI becomes capable of grading safety of new candidates

• Increases accuracy and efficacy by reducing failure

Business rate of animal-to-human trial transition from 90% to 50%
impact (currently 80-90% accuracy)
• Eliminate ethical challenges of animal testing ML

Current user(s) AI provider

User and
provider Digitized cellular Predicted organ-
response specific drug safety
1) also includes “Lead selection”; 2) external data sources are both publicly available and proprietary; 3) organs are created on the chip using human stem cells evolved to mimic the different types of cells in various organs
Source: company website; company reports; EY-Parthenon analysis
Drug development - “Reinforcement learning” (1/1)

Reinforcement learning to leverage existing laboratory books to accelerate and improve

drug discovery steps like NIVIDA BioNeMo
39 Case example: Drug development - “Reinforcement learning” USE CASE | ILLUSTRATIVE

Reinforcement learning principles Implementation prerequisites

► Reinforcement learning is a special type of Required skillset

machine learning based on reward-maximization ► Expertise in natural language processing, deep
underlying function, where each “Right” answer learning and molecular biology, as well as
generates reward and “Wrong” answer – costs familiarity with the SMILES notation for chemical
► The model learns over time by changing actions to structures and the FASTA sequence strings for
maximize the cumulative reward (or minimize amino acids and nucleic acids
costs) Technical requirements
► High-end computing capacities (in-house
GenAI could be used with
Generative model
0c (cc1cc2) ccc1cc2N solutions) or access to cloud-based solutions reinforcement learning to
(external services) to store and process the data generate novel molecules
Stack
► Availability of large and high-quality datasets of with desired properties and
GRU
biomolecular sequences pro properties, such as evaluate their binding
G protein structures, functions, interactions and drug affinity to target protein
<START>
Generated targets
Reward
SMILES Cloud based solutions
Predictive model
Property
► NVIDIA BioNeMo and Google DeepMind are the
leaders in GenAI application in Pharma
► Cloud based solutions allow Pharma companies to
virtually simulate candidates for drugs, which
P c1ccccc1
currently doesn’t exist
Parameter optimization

Source: EY-Parthenon analysis

Return back to meta data

Intelligent dental assistant (1/1)

An intelligent dental assistant can save 30-40% of physician’s efforts, by accurately

diagnosing, recommending personalized treatments & designing optimal dental artefacts
40 Case example: Intelligent dental assistant USE CASE | ILLUSTRATIVE

Intelligent Dental assistant Deep dive – system set-up

Dental physicians often spend a lot of time to analyze Customer asks question Responses can include:
Problem dental images in X-rays, CT scans, leading to errors in in his own words to LLM
► Diagnosis of dental
summarizing insights and formulating treatment plans chatbot images & insights
► Compares different
Additional instructions images of same
Intelligent dental assistant, which can accurately Introduction of additional patient
analyze dental images, summarize dental reports, logic for assisting dental ► 3D dental scenarios
Solution
provide interactive dental education, and generate physicians, e.g., dental training
optimal dental artefact designs condition research, similar ► Generate optimal
treatments undertaken, etc. dental designs
Data retriever
► 30-40% savings in dental physician’s time and
Realized
efforts
business What is the What is the
► Faster turnaround time and improved quality in How can we design What did patients
impact accurate dental effective treatment
diagnosing & treating various dental conditions personalized dental with similar
condition of for patient’s
implants for patient conditions do?
patient? condition?
Impacted ► Radiology
New business
business ► Sales & marketing
models
Dental manufacturing PIM, product Basket analysis,
functions ►
master data, Dental artefacts Customer data,
recommender
product manuals inventory data historic purchases
system
Refer to data quality case
Solutions Information retrieved from various data bases

Source: EY-Parthenon project expertise

Return back to meta data

AI supported NRM (1/1)

AI-aided pricing, discounts and promotions system can increase sales by optimizing
product pricing per customer while achieving higher marketing efficiency
41 Case example: AI supported NRM USE CASE | ILLUSTRATIVE

Optimized NRM1 system Deep dive – AI supported NRM

Intense competition and high degree of price
transparency in (online) retail leave CPG companies Optimizing the Pricing Promotions
Problem struggling to increase revenues and in existing interacting NRM triple:
markets while also experiencing decreasing brand Discounts
loyalty
Required input data (best case):
Optimized pricing and customized discounts / Customer data Product data Market data Consumer data
Solution promotions (NRM) for different customers and online ►
►
Contract data
Sales data
►
►
Price history
Promotion history
►
►
Competitor pricing
Economic data
►
►
Behavioral data
Transactional data
shoppers through AI to increase overall LTV ► Promotion data ► Sales volumes ► Seasonal data ► Demographic
► Counterparts ► Volume forecasts ► Raw material costs information
► Margins ► Package sizes ► … ► …
► … ► …
► Individualmarket / customer pricing strategy
Realized
increasing overall revenues
business
► Improved promotional budget allocation and
impact marketing efficiency Combining, processing
and analyzing data input

Determining ideal promotion Determining optimal price

Sales & Standardized AI
/ discount option range corridors per customer
Key Account, operational
Marketing efficiency
► Optimizedand intelligent pricing, discount &
promotion system per customer / shopper
Solutions
► Effective allocation of promotional budget
1. NRM = Net Revenue Management
Source: EY-Parthenon project expertise

Return back to meta data

Accounting Analytics (1/2)

AI-powered analytics tool that uncovers Accounting issues and anomalies and suggest
mitigating actions based on large history of accounting data
42 Case example: Accounting Analytics USE CASE | ILLUSTRATIVE

Accounting Network Analytics powered by GenAI Deep dive – Accounting Network Analytics

Starting point: List Transformation Network for one Network for the
Accounting anomalies are hard to discover as they into list of edges specific journal entry entire general ledger
of journal entries
are often obscured due to the format and tabular set-
Problem JE ID: 100231
up of the general ledger. This leads to inefficiencies From:
Receivables
and pose a risk for the quality of the Close and Audit. To: Sales
Amount: 100

The accounting network transforms the general ledger

into a graph, enabling finance professionals to obtain a
Solution complete overview of complex accounting practices
quickly and intuitively. Comprehensive algorithms add
additional insights.

► Ambiguous value flows are made transparent (e.g.

incorrect JE postings, unnecessary transactions,
Realized etc.) and
business ► Increased data quality
impact ► Documentation of effectiveness of internal controls
► Increased efficiency in accounting operations

Solutions

Return back to meta data

Accounting Analytics (2/2)

Key findings from the application at EY’s core audit clients

42 Case example: Accounting Analytics USE CASE | ILLUSTRATIVE

The excessive use of Offsetting payables and receivables is
Hidden bookkeeping
transit accounts only allowed under rare circumstances.
opens the door for The posting of payables to receivables
all types of may represent a GAAP violation and a
undesired behavior. dangerous pattern for masking
It undermines any receivables from fictitious revenue.
chance for review Incorrect
journal entry
and possibly violates
posting
GAAP. We call it “the
black hole” in
bookkeeping. Easy
catch for fraudsters.

Incomplete mapping

Ambiguous journal entries Unnecessary transactions Circular

Ambiguous journal entries
transactions,
(with more than two debit
transactions with
and more than two credit
zero amount and
accounts) add an
other unnecessary Account classes with significant
unnecessary complexity to
transactions – we transaction volume labeled “not
bookkeeping. They may
have seen used” or simply not mapped to
result from “working
companies where any balance sheet or income
backwards” from the
one third of journal account class.
account movements
entries was useless.
intended with a posting
rather than the real
economic event.
Transaction Anomaly Detection – Procure to Pay (1/1)

Introduction of a Anomaly Detection AI can result in savings in accounts payables,

reduces cycle times by straight through processing and minimizes operational risk
43 Case example: Transaction Anomaly Detection – Procure to Pay USE CASE | ILLUSTRATIVE

Input: Data Sources EY Transaction Anomaly Detection AI Key Results

Current Vendor Invoice Details Fraud Detection

(e.g., invoice item wise current The solution leverages AI techniques such as anomaly (e.g., enables early detection of fraud hence
outstanding) limiting revenue leakage)
detection and pattern recognition to analyze
Past Vendor Invoice & PO transactions to look for excess or duplicate payments, Erroneous Transactions
(e.g., invoice trends, transactional possible maker-checker collusion, payments to shell (e.g., flags transactions which does not
invoice data etc.) correspond to past trends)
vendors, suspicious vendor invoices, etc. resulting in
identification of outliers and fraudulent transactions
Vendor Demographics Auto retrain based on user feedback
(e.g., vendor profile, location, other (e.g., human reinforced learning to retrain AI)
details from CRMs)

Application Duplicate PO to Invoice

Suspicious Invoices Insights Engine Excess Payments
Integration Payments Discrepancy

Identifies
Identifies Identifies cases
Integrate with single discrepancies
transactions not where excess
or multiple ERPs, Identifies potential Based on past between PO and
inline with past payments are being
CRMs through APIs, maker-checker payment, identifies invoice which are
trends both at made before the
database connectors collusion duplicate payments not inline with
invoice and at line transactions actually
etc. historic trend of
items level occur
exceptions

Powered by EY ASpace AI Platform

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

Automation of EBIT-Forecasting (1/1)

A leading public infrastructure company was able to improve the EBIT forecasting
accuracy by 36% by applying Machine Learning based forecasts
44 Case example: Automation of EBIT-Forecasting USE CASE | ILLUSTRATIVE

Client Challenge EY Approach

► Past forecasts for relevant KPIs were historically inaccurate ► Implementationof a Machine Learning based forecast with automatic
► Inaccuracy of KPIs (e.g. EBIT) let to mistrust, why KPIs were not used as Hyperparameter-selection
part of the financial planning ► Data-Validation and analysis of the current accuracy of running forecasts
► Missing data complicated the implementation of a rolling forecast ► Integrationof Business Rules based on segment-specific behaviours to
► Thus, limited reliability was given on which months / deviations from improve the forecasting accuracy
plans were leading to the overall deviation ► Development of a data-model for a scenario-based planning
► Historical deviation could thereby not be excluded from future ► Operationalisation of dashboards for planning and analysis purposes
forecasts, leading to unreliable forecasts ► Analysis of more than 130.000 data points

Value created by solution

36% more accurate in 52 Positions were Self-learning system, Implementation of a Month over Month
EBIT-forecasts automated that improves by time rolling forecast and agile visualization and
based on business rules planning cycles enabling forecasts enabling fast
and external factors fast adjustments to adjustments in case of
markets changes planning mistakes

Source: EY-Parthenon project expertise

Return back to meta data

AI Powered AR Collections Assistant (1/2)

Introduction of AR Collection Assistant can lead to 30% improvement in DSO, 40%

improvement in productivity and 22% improvement in Receivable period (1/2)
45 Case example: AI Powered AR Collections Assistant for Finance USE CASE | ILLUSTRATIVE

Input: Data Sources EY AR Collections Assistant Key Results

Payment Information AI Assist

(e.g., invoice item wise current (e.g., enabling collections strategy,
outstanding) conversation recommendations etc.)
An intelligent voice enabled assistant with advanced NLP
Past Customer Trends and ML capabilities to help improve collections rate, Improved Metrics & Cash Flows
(e.g., credit, payment trends, (e.g., increase in receivables, collections rate
demographics, disputes) reduce DSO and thus leading to improved cash flows and reduction in DSO)

Discussion Details Single Glass View

(e.g., notes from past discussions, (e.g., one window to update and view any
promise to pay ) information)

Voice Enabled ML based worklist AI to Identify ‘At ‘Next Best Action’ Auto Dunning
Data Ingestion
Assistant prioritization Risk’ Customers AI Engine e-mailers

Natural Language Anomaly detection Based on past

Leverage Machine
Processing and and Pattern customer profile and
Prebuilt connectors Learning (ML) to AI Assistant can be
Entity Extraction to Recognition payment trends,
through API with prioritize accounts asked to send Auto
retrieve customer leverage past recommend
any number of CRM and invoice with dunning emailers to
information and customer payment personalized Next
or ERPs whom follow ups eligible customers
auto update patterns and Best Action for
needs to be done
ERP,CRM interaction history every customer

Powered by EY ASpace AI Platform

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

AI Powered AR Collections Assistant (2/2)

Introduction of AR Collection Assistant can lead to 30% improvement in DSO, 40%

improvement in productivity and 22% improvement in receivable period (2/2)
45 Case example: AI Powered AR Collections Assistant for Finance USE CASE | ILLUSTRATIVE

Client Challenge Key Features Key highlights

► A fortune 100 global manufacturing company was facing a challenge

where the Credit Analysts (CA) from Cash to Collections (C2C) team
requires to call customers to follow up for payment. 20% Improvement in
► There was no recommendations available to the CA regarding next best DSO
action, no customer profiling, no analytics regarding customer history and Next Best Action
multiple SAP updates which was time consuming. Recommendation
► No transparency regarding effective CA time utilization 2% Improvement
collections receivables
rate

Solution Overview Risk Classification

► A Space conversation and machine learning module was leveraged to 63% Improvement in
implement a chatbot on premise. The solution provides a voice enabled data accuracy
chatbot which is integrated real-time with ERP applications and skype for
business which enables a single view interaction window for the agent to Anomaly detection
view and update information. 40% Improvement
► In addition, machine learning was leveraged to classify customers into risk Agent productivity
category based on past payments, proactively detect anomalies in
payments and suggest next best action based on machine learning enabled
Realtime
pattern recognition Application
Integration

Source: Desk research; EY-Parthenon project expertise

Chatbot for [IFRS 9] and internal accounting standards (1/1)

Chatbot based solution supports the consistent and fast application of correct
accounting standards – combining current [IFRS 9] and internal accounting standards
46 Chatbot for [IFRS 9] and internal accounting standards USE CASE | ILLUSTRATIVE

IFRS 9 Chat Assistant – Process Structure

Refomulate
answer

Yes
No
Was similar answer
already reviewed?
No
No Retrieve rel.
User asks Summarize Generate
document Get answer Yes
question question promt
chunks
Yes No
Retrieve
Return Ask follow up
question and
answer question
answer Yes
Does this answer
my question?

Value created by solution

Unification of IFRS 9 and Reduction of research Reduction of knowledge Faster and specific
Golden Source
internal standards and interpretation time monopolies knowledge consumption
LLM Action User Action Cognitive Search Backend Action
Source: EY-Parthenon project expertise

Return back to meta data

AI-Enabled reports processing vs manual alternative (1/1)

Auto commentary is a very attractive use case to show the value potential of GenAI

47 Use case example: AI-Enabled reports processing vs manual alternative USE CASE | ILLUSTRATIVE
Input Efficiency Efficiency/Insight Insights

Step 1: PoC for Step 2: Non-standard Step 3: Deep root

standard reports reports cause analysis
GenAI maturity for commentary
of financial reports
► “From database "Private learning" capabilities to
► AI generated pre- retrieve the gap make results client-specific and
commentary to between planned and ► “Give me products ensure confidentiality
support controllers in forecasted revenue which will probably Enablement of explainability of
identifying outliers, based on the forecast decline over next 6 complex analysis results
AI-Enabled for January over the
drivers, etc. months, and provide
► Controllers use pre- last 10 years per reasoning why.“ Benefits expected with GPT-3.5
comment as a draft, product segment Improvements expected with
Standard
validate findings and ► Afterwards write GPT-4 across steps (e.g., from
reports small summary of enhanced numerical capabilities) –
refine comments
rationales to explain EU availability given, pricing the
historic deviation for same according to Microsoft
this year per
segment“

► Initial manual query Working hypothesis:

► Manual one-time Not feasible at scale
by expert manual & ► PoC and evolutionary use cases to
query & case-by-case
Alternative time consuming case- ► (expert required) be implemented on top of current
investigation
by-case investigation reporting landscape

Source: EY-Parthenon analysis

Return back to meta data

Translation service (1/1)
GenAI powered translation services use natural language processing algorithms to translate
documents and data files while preserving the original formatting
48 Use case example: Translation service USE CASE | ILLUSTRATIVE

Translation service How it works?

► Time-consuming process of accurately translating Translate into desired

large amounts of client requests or ad content into language
different languages and formatting challenges
Problem
► Various markets cannot be addressed in a common
approach due to language barriers (e.g. Eastern
Europe) Marketing & sales Customer Ad creation Customer care
content creation onboarding content request

► GenAI powered language translation and

formatting can efficiently translate content from one
Customer request is processed by (Gen)AI by leveraging marketing
Solution language to another while preserving the original
formatting and sales material, product manuals, ad content and client data
► Leveraging semantic translation capabilities of GenAI

Language trans-
formed in GenAI GenAI
processable prompts

Realized ► Fasternew market entries and go-to-market

Ad campaign translation
business ► Enhanced customer service
with clear, unaltered, and
impact ► Improved efficiency of campaign creation
congruent output into
Response
text, visuals or voice

Source: EY-Parthenon project expertise

Return back to meta data

GitHub Copilot (1/1) Use case category:

Microsoft’s GitHub Copilot is an AI-powered coding assistant that increases L4: Implemented with client
L3: Built for EY

developers’ productivity and helps bridge skill gaps L2: Built as a Prototype
L1: Proof of concept

49 Case example: GitHub Copilot USE CASE | ILLUSTRATIVE

GitHub Copilot Demo – GitHub Copilot in Python1

► GitHub Copilot is an AI-powered coding assistant

developed by GitHub in collaboration with OpenAI Task: “Write a function to validate
that runs on a GPT 3 language model if a password is strong
Description ► Helps developers write code more efficiently by
providing code suggestions (in multiple
programming languages) and generating code based
on natural language descriptions
► Provides a comprehensive coding solution by offering
context-aware code suggestions, automatically
Solution converting comments into code, autofilling
repetitive code segments, and presenting alternative
code options for enhanced development productivity

Realized ► Increases developer productivity

business ► Accelerate innovation, e.g. through rapid
impact prototyping and innovation and bridge skill gaps

Impacted Finance & Controlling

Operational
business Sales & Marketing
efficiency
IT & Technology
functions
Context
Solutions Suggestions
Exemplary; Github Copilot provides coding support across multiple coding languages
Source: EY-Parthenon project expertise

Return back to meta data

Code generation (1/1)
Generative AI can drastically shorten software development time & efforts, transform data
generation, processing, & automate visualizations using natural language prompts
50 Use case example: Code generation USE CASE | ILLUSTRATIVE

Code generation How it works?

► Increased time to deliver software – Developing

software in traditional ways takes a lot of time, and
 Text

“ GitHub works might involve developing a feature that uses AI to enhance

code collaboration and management on the GitHub platform.

Problem software code quality is dependent on skill set of Intelligent code suggestions Rapid Code Generation Enhanced Code Understanding
► It analyzes the context and code ► Copilot can quickly ► by providing contextual
developers in real-time to generate accurate generate repetitive code suggestions and
and relevant suggestions, saving snippets based on explanations to unfamiliar
developers time by automating developer's input and code bases. It offers insights
repetitive coding tasks information it has learned into API usage, offers
from a vast codebase alternative implementations

► Gen AI can automate complex coding via natural

language prompts (code snippets, functions, coding
Solution suggestions, assist in identifying bugs, refactor / Commits Code
Integration with code editors and version control
translate code), automate testing of various
scenarios

Optimization criteria:
► Enhanced Code quality
► Increased Developer
Productivity
► Enhanced security
Realized ► Significantsoftware engineering cost, effort ► Enhance automation

business and time savings

impact ► Faster time to value of software solutions

Source: EY-Parthenon project expertise

Return back to meta data

Customer Care Service (1/4) Use case category:

GenAI can help optimizing customer care service via a Virtual Assistant that either
L4: Implemented with client
L3: Built for EY

solves client queries in full or helps human agents augmenting their capabilities L2: Built as a Prototype
L1: Proof of concept

x Case example: Customer Care Service – VA for incoming customer calls USE CASE | ILLUSTRATIVE
Virtual Assistant for incoming customer calls Deep dive – Our Approach

 The client, one of Spain’s largest telecommunications

Problem providers, was handling a huge amount of customer call
queries per day exceeding agents’ capacities resulting in
an inefficient service and client dissatisfaction

 As part of a comprehensive transformation of the

customer service operating model utilizing traditional and
generative AI, the creation of a ChatGPT-boosted VA was
accomplished
 The VA works as a first contact in a customer’s call.
Solution  It extracts relevant information from the conversation
identifying the client, the product or service involved and
the call’s intent so that, whenever the call is not resolved
by the VA, it is redirected to a human agent providing an
accurate summary of the conversation and suggested next
steps easing resolution

The project attained the following results:

Realized  30% reduction in average call duration
business impact  45% reduction in unsatisfied customers
 25% OPEX reduction

Source: EY-Parthenon project expertise

Page 167 20 August 2024
Customer Care Service (2/4) Use case category:

GenAI can help optimizing customer care service via a Virtual Assistant that either
L4: Implemented with client
L3: Built for EY

solves client queries in full or helps human agents augmenting their capabilities L2: Built as a Prototype
L1: Proof of concept

x Case example: Customer Care Service – VA for incoming customer mails USE CASE | ILLUSTRATIVE
Virtual Assistant for incoming customer calls Deep dive – Our Approach

 The client, one of Spain’s largest telecommunications

Problem providers, was handling a huge amount of customer mail
queries per day exceeding agents’ capacities resulting in
an inefficient service and client dissatisfaction

 As part of a comprehensive transformation of the

customer service operating model utilizing traditional and
generative AI, the creation of a ChatGPT-boosted VA was
accomplished
Solution
 The VA is able to rank and classify customer mails, it
categorizes the customer as well as the intent and scope
of the mail and suggests an automated reply that can be
quickly confirmed by the human agent

The project attained the following results:

Realized  30% reduction in average call duration
business impact  45% reduction in unsatisfied customers
 25% OPEX reduction

Source: EY-Parthenon project expertise

Page 168 20 August 2024 Return back to meta data
PPE pre-emptive maintenance (1/3) Use case category:

Gen AI can monitor PPE functioning, provide live deterioration forecasts improving
L4: Implemented with client
L3: Built for EY

assets’ pre-emptive maintenance and significantly reducing the risks of breakdown L2: Built as a Prototype
L1: Proof of concept

x Case example: PPE pre-emptive maintenance - Rail Network Operator USE CASE | ILLUSTRATIVE
Rail track pre-emptive maintenance

 The opportunity to use data and analytics to accurately identify track locations likely to be impacted by temporary speed restrictions (TSRs) and
geometry defects is key to effective long-term planning and intervention measures. Insights at speed solve these common user pain points: a)
Problem Forecast TSRs that have the most impact on cycle time to inform intervention activity in the short and long-term; b) Pre-emptively identify areas
along the rail network that are prone to geometry defects or deterioration to effectively prioritize maintenance or intervene to prevent TSRs from
occurring

 The main solutions developed for tackling the complex client’s problematic are a) Track Forecasting Tool (TFT): visualize and forecast long-term
track health conditions, maintenance plans and generate intervention scenarios. b) Maintenance Impact Model (MIM): Simulate and forecast
Solution improvement in geometry after tamping intervention; and c) Plan Performance Assessor (PPA): The PPA reports key metrics used in the
assessment of executed maintenance interventions. The historical metrics visualize the change between the condition or performance of the
network before and after the intervention

The project attained the following outcomes:

 35% faster cycle time (mine to port return) *57 min improvement
Realized
 21% faster velocity (average speed of travel) *5.9km/h increase in average velocity
business impact
 45% fewer TSRs on the network *TSRs reduced by 65
 33% healthier Assets (reduction in serious E1 issues) *90+kms of E1s reduced to only ~60kms

Source: EY-Parthenon project expertise

Page 169 20 August 2024 Return back to meta data
PPE pre-emptive maintenance (2/3) Use case category:

Gen AI can monitor PPE functioning, provide live deterioration forecasts improving
L4: Implemented with client
L3: Built for EY

assets’ pre-emptive maintenance and significantly reducing the risks of breakdown L2: Built as a Prototype
L1: Proof of concept

x Case example: PPE pre-emptive maintenance – Structural Health Monitoring System USE CASE | ILLUSTRATIVE
Structural Health Monitoring tool Deep dive – Our Approach

 A structural health monitoring (SHM) system is defined as

the use of on-structure sensing system to monitor the
performance of the structure and evaluate its health
Problem state
 The client structure was a 300-year expected life bridge
that carries over 110K vehicles daily volume making it
prone to breakdowns

 A SHM has been installed on a 71-metre span on the old

bridge with 60+ sensors packages. The SHM monitors the
Solution structural integrity of the span with predictive damage
detection capabilities providing decision support for
intrastructural management

 Real-time monitoring of the asset performance, alerts,

and visual assistance
Realized  Condition-based maintenance that is aided by predictive
business impact machine learning models
 Improved benefits/cost ratios in asset operations,
prolong asset life and lift operation productivities

Source: EY-Parthenon project expertise

Page 170 20 August 2024 Return back to meta data
PPE pre-emptive maintenance (3/3) Use case category:

Gen AI can monitor PPE functioning, provide live deterioration forecasts improving
L4: Implemented with client
L3: Built for EY

assets’ pre-emptive maintenance and significantly reducing the risks of breakdown L2: Built as a Prototype
L1: Proof of concept

x Case example: PPE pre-emptive maintenance – Production Line Industrial Robots USE CASE | ILLUSTRATIVE
Production Line Industrial Robots Deep dive – Our Approach

 Industry robots, epitomised by the robotic arms, usually

refers to a robot system used for manufacturing.
Industrial robots are automated, programmable and
Problem capable of movement on three or more axes. These
robots represent highly stressed production systems,
and their wear-related breakdown can lead to long
production downtimes and thus to high costs

 A prototype system has been developed to meld together

Digital System Models with Digital Twins for the client’s
robots, used here as surrogates for more advanced robots
Solution  The system enables data-supported system monitoring
and pre-emptive maintenance for systems’ and
components’ health, aiming at optimising maintenance
resourcing by moving from time-based to prioritised
condition-based maintenance

 Condition monitoring, failure prediction, intelligent

diagnosis, and pre-emptive maintenance
 Shift operation from time-based maintenance to
Realized
prioritised condition-based maintenance by using machine
business impact learning models
 Prolong robot life, sustain manufacture productivity,
reduce operational cost

Source: EY-Parthenon project expertise

Page 171 20 August 2024 Return back to meta data
Demand Forecasting (1/2) Use case category:

Gen AI can forecast sales ameliorating production planning, resource allocation in

L4: Implemented with client
L3: Built for EY

advance significantly minimizing costs of unexploited inventories or excess staffing L2: Built as a Prototype
L1: Proof of concept

x Case example: Demand Forecasting - Palm Oil Producer USE CASE | ILLUSTRATIVE
Palm Oil Producer

 A palm oil producer is heavily reliant on crude palm oil (CPO) as the base material, which in turn is used to create by-products such as palm oil or
margarine
Problem  The market demand for both CPO and the refined by products is volatile from both commodity trading and customer consumption perspectives
 The client was producing either far above or far below demand entangling in a vicious circle alternating liquidity loss due to acquisition of excess
inventories and unmet demand losing competitiveness

 The client needed a capability to effectively plan production and forecast sales. This insight enables them to more effectively manage production
cycles and respond quickly to product volatility and inventory to deliver the best outcome for the customer
Solution
 The EY DnA team has built an integrated data platform which gathers operations systems data and applies machine learning to simulate production
and recommend production improvements to better meet demand

Realized  This integrated forecasting, automated alerts and production simulation helps reduce unnecessary inventory, and contributes to the goal of
business impact reducing working capital by $3M

Source: EY-Parthenon project expertise

Page 172 20 August 2024 Return back to meta data
Demand Forecasting (2/2) Use case category:

Gen AI can forecast sales ameliorating production planning, resource allocation in

L4: Implemented with client
L3: Built for EY

advance significantly minimizing costs of unexploited inventories or excess staffing L2: Built as a Prototype
L1: Proof of concept

x Case example: Demand Forecasting - Port Logistics Operator USE CASE | ILLUSTRATIVE
Port Logistics Operator

 Logistics present an overwhelming challenge for one of the world’s largest port operators. Nearly 100 vessels move in and out every day, each
carrying up to 20,000 20-foot containers, requiring giant trucks, cranes and workers with diverse skills to service them. Any miscalculation in a
Challenge vessel’s arrival time has the potential to create significant delays for the port. When this takes place, vessels are left without somewhere to dock,
workers wait without a job to perform, and machinery gets bottlenecked
 Helping the client develop an approach to accurately predict vessel movement and from there, better allocate people and machinery was the remit

 Using AI machine learning and new data sources, EY helped develop three live solutions with more than 20 AI models built and integrated to
predict: vessel arrival 14 days out, resources the port would need to meet demand, and potential machinery choke points in the yard
 The team used extensive historical databases to inform the machine learning model with variables such as tidal patterns, vessel sizes and past
Solution variations in arrival times from different shipping lines. By incorporating this historical data with satellite navigation and each vessel’s declared arrival
time, the vessel movement model proved to be materially more accurate than the port operator’s previous system
 More precise arrival times subsequently allowed for more accurate workforce planning; which, in turn, better highlighted potential choke points and
allowed for real-time suggestions for faster driving routes through the yard so that machinery was where it needed to be at the right time

 Together, these three automated AI solutions significantly increased the port’s efficiency and helped to deliver substantial cost savings. The
automated AI solution for vessel movement alone increased accuracy by 3% which translated to approximately US$10.2 million in savings in the near
Realized term
business impact
 In addition to this macro impact for the client, better predictions have meant that workers on the ground now have more predictable work hours, with
little down-time during a shift and reduced over-time due to unexpected vessel delays

Source: EY-Parthenon project expertise

Page 173 20 August 2024 Return back to meta data
Dynamic Pricing Platform (1/1) Use case category:

Gen AI can boost dynamic pricing mechanisms delivering real time personalized
L4: Implemented with client
L3: Built for EY

prices based on customer profile and behavior patterns enhancing customer L2: Built as a Prototype
L1: Proof of concept
engagement
x Case example: Dynamic Pricing Platform USE CASE | ILLUSTRATIVE
Dynamic Pricing Platform for an insurance firm

 The client was an insurer experiencing challenges around the complexity of their IT and systems environment, with multiple premium pricing
Problem engines and over 57 core legacy policy and claims systems leading to fragmented data management practices impeding their ability to target
customer needs to better serve these and maximize value through pricing

 EY’s DnA team conducted a pricing transformation through a pricing engine algorithm simplification
 Uplifted pricing capability through implementing a cloud-based real-time pricing platform
Solution
 The team supported the training of new pricing machine-learning models and transitioning the client’s pricing process onto the enterprise data
platform

 Alignment with future cloud vision: The new platforms and data solutions deliver on the client’s cloud transformation vision
 Data driven pricing and enhanced customer engagement: The new pricing platform will deliver real-time personalised pricing based on customer
Realized profile, behaviour patterns etc, enhancing the customer experience
business impact
 Operational efficiency: Harmonisation of reports across all business functions across AU/NZ leading to huge operational efficiency outcomes
 Real-time decision-making capability for the first time

Source: EY-Parthenon project expertise

Page 174 20 August 2024 Return back to meta data
Network usage decision-making (1/1) Use case category:

Generative AI can help improving network usage decision-making for the day-to-
L4: Implemented with client
L3: Built for EY

day operations L2: Built as a Prototype

L1: Proof of concept

1 Case example: Network usage decision-making - Rail track priorization USE CASE | ILLUSTRATIVE
Rail track priorization

 The client has thousands of kilometres of rail network carrying trains on a continuous loop of more than 20,000 kms travelling between mines and
Problem ports every day. Bottlenecks across the network, track degradation and maintenance activities was costing the business millions of dollars in
downtime, revenue, resourcing and asset performance

 EY teams collaborated with the client and subject matter experts to develop an artificial intelligence, deep machine learning solution that predicts
Solution and prioritises rail track condition issues and provides day-to-day optimal path usage recommendations that returned the highest business value
 The Al machine learning models predict whether a problem will become a network constraint as well as its associated business impact

The project attained the following outcomes:

Realized  A solution that could help to identify which issues were impacting the bottom line the most
business impact  Fix underlying track problems and remove rail track issues safely and quickly
 Reducing time off the mine to port return journey, which is translated into savings to the client every day

Source: EY-Parthenon project expertise

Page 175 20 August 2024 Return back to meta data
Text digitization by Optical Character Recognition (1/2) Use case category:

OCR tools are the most versatile, scalable GenAI application to business processes
L4: Implemented with client
L3: Built for EY

enabling text digitization vastly augmenting traceability, organization of documents L2: Built as a Prototype
L1: Proof of concept

x Case example: Text digitization by OCR – Income and expense verification for Investment Bank USE CASE | ILLUSTRATIVE
AI-automated income and expense verification process for a Major Australian bank

 The current loan application process requires a large team of employees to manually verify and validate over 1,000 applications per day, equating to
thousands of pages across multiple different types of documents and formats (e.g. payslips, tax returns, and bank statements as photocopies,
PDFs, images, etc) These are income and expense documents, with each application taking several hours to verify, resulting in significant
Challenge operational overhead, human error, slow loan approval times, and low customer satisfaction
 A modern, scalable solution was required to automate the process, improve application times, achieve operational efficiency and enhance the
customer journey

 A new, Azure-based artificial intelligence solution now supports the income and expense verification process for loan applications
 Combining Microsoft’s AI Cognitive Services including Azure Form Recognizer and Computer Vision (OCR) and EY’s custom natural language
processing modules, the solution achieves high levels of precision at classification and high levels of accuracy of extraction for the key fields
Solution
contained in those documents, such as base salary, address, and payslip frequency. The results are then displayed neatly in a web-based UI
 The EY team also added some additional fraud detection capability to the solution; by cross-validating income across pay slips, bank statements, and
ATO NOA (Tax Documents), further strengthening the value of the solution

 Reduced operating expenditure with automation of the end-to-end process

 Improved regulatory compliance and risk mitigation with fraud detection and data surveillance built into the solution
 Better customer and employee experiences with a simplified process requiring minimal interactions, and reduced approval times
Realized
 Greater and faster accuracy of document classification and verification of key fields saving thousands of full-time employee (FTE) days per year
business impact
through intelligent automation
 Flexibility to scale up or down to meet demand, built into the technology environment, or delivered as a service. The solution is estimated to
continue to process larger volumes of data over time

Source: EY-Parthenon project expertise

Page 176 20 August 2024 Return back to meta data
Text digitization by Optical Character Recognition (2/2) Use case category:

OCR tools are the most versatile, scalable GenAI application to business processes
L4: Implemented with client
L3: Built for EY

enabling text digitization vastly augmenting traceability, organization of documents L2: Built as a Prototype
L1: Proof of concept

x Case example: Text digitization by OCR – Patient questionnaire recording for a hospital USE CASE | ILLUSTRATIVE
AI-automated patient questionnaire processing and cloud load

 The key challenge of one of Australia’s biggest hospitals is managing large volumes of manual patient questionnaire processing that come through
and the ability to process them in a timely manner so the outputs can be used to drive effective decision making
Challenge  At peak times, 100+ people are solely responsible for taking hard copy questionnaires and lab documents from the infectious disease agency and
entering that data into an online system. The lack of a digital workflow is inefficient, locks critical resources to data entry tasks, processing that is
prone to human error and requires people to work on site, causing health and safety risks

 EY teams designed a new workflow that digitizes hard copy documents and processes them using machine learning, AI-driven optical character
recognition (OCR) that converts documents, extracts the information within it and loads the data into the system
Solution  A new system has been built to support the new workflow, which conforms to the strict data privacy, information security and auditing policies. The
teams conducted in-depth studies on workflows and underwent intensive information security protocols and a comprehensive solution vetting
process

The new workflow delivers these results:

 Bulk document and data upload capability

Realized  Reduced document error

business impact  Unlocking resources away from manual data entry to value-added tasks such as data quality, epidemiology and frontline response
 Easily maintained process, no AI or data and analytics technical expertise required by the staff
 Seamless integration with existing systems

Source: EY-Parthenon project expertise

Page 177 20 August 2024 Return back to meta data
Reconciliation process for Investment Bank (1/2) Use case category:

Generative AI can help financial entities establishing the desired rules-based

L4: Implemented with client
L3: Built for EY

systems in machine learning models to automate their reconciliation processes L2: Built as a Prototype
L1: Proof of concept

x Case example: Reconciliation process for Investment Bank USE CASE | ILLUSTRATIVE
Machine learning reconciliation for an investment bank

The client’s reconciliation process was proving insufficient with the existing rules-based system and, as a result, a large operations team was manually
processing breaks. Key challenges included:
 A high number of breaks daily that could not be effectively managed manually
Problem
 The existing reconciliation tool was unable to manage multiple matches (i.e. consolidated entries)
 No prioritization or closest match provided by the rules-based reconciliations tool
 Significant FTE effort spent each day to find matches that could be spent on higher value work

 The EY DnA team built a machine learning model which has improved the reconciliations process by providing recommended matches to the
operations team on a day-to-day basis
Solution  The new model was designed to be more flexible in handling breaks and make recommendations for subsequent human action, accelerating the
overall process as a cognitive aid
 The intention is to productionise the model to enhance the existing rules-based system that had proved to be insufficient

 The ML model is expected to find matches up to 50% quicker than the previous manual process, as well as to provide a prioritization to further
accelerate the matching process. We saw a 60% match rate on all complete matches for a given day
 This will significantly reduce the FTE effort spent each day on manually identifying matches. The prioritization of cases will also allow the team to
Realized provide more accurate figures to senior management on the true number of remaining breaks
business impact
 The solution was developed in a scalable fashion that will allow it to be rapidly deployed for other reconciliations processes, and be productionised
 Analysis of the machine learning model results allowed us to provide the client with insights into potential enhancements for the matching logic in
their current rules-based reconciliations tool

Source: EY-Parthenon project expertise

Page 178 20 August 2024 Return back to meta data
Reconciliation process for Investment Bank (2/2) Use case category:

Generative AI can help financial entities establishing the desired rules-based

L4: Implemented with client
L3: Built for EY

systems in machine learning models to automate their reconciliation processes L2: Built as a Prototype
L1: Proof of concept

x Case example: Reconciliation process for Investment Bank USE CASE | ILLUSTRATIVE
Deep dive - Machine learning reconciliation for an investment bank

Page 179 20 August 2024

Customer Credit Risk Management (1/1) Use case category:

GenAI models can help financial entities managing customer credit risk by
L4: Implemented with client
L3: Built for EY

providing accurate client default propensities L2: Built as a Prototype

L1: Proof of concept

x Case example: Customer Credit Risk Management USE CASE | ILLUSTRATIVE

Customer Credit Risk Management for a Credit Union

 The client was one of Australia’s major credit unions managing the accounts of thousands of households. With credit risk management being the key
Problem issue to sustain the union, they counted with an out-of-date tool for targeting client’s risk-adoption patterns which was leading to fund
mismanagement regarding their mortgages’ services (which account for 40% of the union’s services)

 The EY DnA team assisted the financial service provider to manage consumer mortgage risk by identifying cross-sell opportunities and help drive
targeted retention initiatives for customers with high propensity to disengage. Data science models generate the propensity of the customers to
default allowing the client to drive targeted intervention
Solution
 The team supports the cloud-based solution to ingest data from the client’s ERP, run the statistical models and generates the output models for
downstream consumption on a daily basis. The solution components include data science models to predict customer churn, cross sell and upsell
potential and the maintenance of data security and integrity

Realized  Improvement in the accuracy of the risk assessment process by enhancing the previously existing tool by the ML models
business impact  EY DnA selected as a finalist in the “Engage Your Customers” category of the Microsoft Australian Partner Awards

Source: EY-Parthenon project expertise

Page 180 20 August 2024 Return back to meta data
Identification of cartographic areas of interest (1/1) Use case category:

The combination of GenAI with satellite imagery can help governments, as well as
L4: Implemented with client
L3: Built for EY

land-intensive businesses, target cartographic areas of interest L2: Built as a Prototype

L1: Proof of concept

x Case example: Identification of cartographic areas of interest USE CASE | ILLUSTRATIVE

Identification of cartographic areas for a government’s Department of Defense

 Australia’s National Unexploded Ordnance Program (NUXOP) is responsible for identifying, managing and remediating unexploded ordnance (UXO)
that poses a risk to public safety
Problem  However, accurately identifying the location of contaminated sites is a major challenge for managing the risk of UXO
 There is a need for more efficient methods of identifying the location of potential UXO to enable the safe and effective management of UXO on
Defence estates and to comply with Commonwealth Policy

EY's team worked closely with client subject matter resources to develop a customized solution to address the special characteristics (including activity
status and vegetation coverage) of each site in scope for potential UXO identification, using a combination of optical satellite imagery and secondary
open-source data. The solution included the following key components:
Solution  Image processing: Analysing and manipulating various spectral bands to enhance visual signals of crater-like features
 Machine learning: Separating crater-like features from the background when there is a clear visibility of the ground surface
 Image segmentation and classification: Identifying areas of interest (AOI) for further investigation based on identifiable features
 GIS suitability analysis: Applying a ranking and scoring system to a number of GIS features to detect high-interest areas for UXO

 The identification of areas of interest in scoped sites for further investigation of potential UXO, which can help to reduce the risk to public safety
 Proven use of optical satellite imagery, including machine learning and GIS modelling, to provide effective and efficient methods of identifying
Realized
potential UXO
business impact
 Comparison to other technologies (outcome, time and cost), which can help to improve the accuracy and reliability of UXO identification and inform
decisions about future technology adoption and investment

Source: EY-Parthenon project expertise

Page 181 20 August 2024 Return back to meta data
Accounting and Reporting automation (1/1) Use case category:

GenAI can help automating asset cost determination and capitalization processes
L4: Implemented with client
L3: Built for EY

by producing highly visual predictive and traditional reporting L2: Built as a Prototype
L1: Proof of concept

x Case example: Accounting and Reporting automation – Fixed Asset Cost and Capitalization USE CASE | ILLUSTRATIVE
Fixed Asset Cost and Capitalization Reporting Automation

 The management of the client’s fixed assets capitalisation procedures and registers was highly manual and prone to mistakes, resulting in a high
level of financial risk but at the same time was draining resources that could be deployed elsewhere
 The client required a robust, technology enabled system that analysed their cost and made tax and accounting determinations without manual
Challenge
intervention but was also able to produce accurate and transparent results
 They also sought a platform that could easily produce relevant reports that could be utilised for budgeting and financial management purposes such
as dashboards that reported on the asset base, forecasting and predictive modelling of their portfolio

 The EY Elementary Delivery Team’s asset register operates in a secure web and cloud-based environment, allowing the relevant members of the
client’s Finance team to securely connect to both the EYE register and AI Centre platforms using various devices. Users can generate various tax
and accounting depreciation reports at any time, including the forecasting of future depreciation which can be used for budgeting and
management purposes
Solution  The platform also provides the client with various fixed asset reporting and analytics dashboard capabilities
 The classification of capitalised costs was reviewed by the Elementary AI Centre module which applies machine learning and EY industry
knowledge processes. The solution’s key components are: SAP processed descriptions, invoices and purchase orders for all sample assets,
recognition of expenditure type, correct allocation of cost attributes concerning depreciation methods and relevant invoice documentation and
automatic cloud upload

 EY successfully delivered the automation of fixed assets’ cost determination and capitalisation processes, the visualisation of its asset base and
producing highly visual predictive and traditional reporting
Realized
 Reducing exposure to financial risk and integrity
business impact
 Significant increase in productivity for the client due to its transition from a highly manual processing and review environment to a fully automated
environment

Source: EY-Parthenon project expertise

Page 182 20 August 2024 Return back to meta data
GenAI
use cases (additional
slides – demo video
based, and other use
case)
 Use case category:

Company sentiment analysis - generates the future outlook of a company

L4: Implemented with client
L3: Built for EY

based on its annual financial report and annual sustainability report

L2: Built as a Prototype
L1: Still conceptual

12 Case example: Company sentiment analysis

Use Case
The client wishes to set up a company sentiment analysis tool for richer insights generation on future outlook, planning and enhancing company’s financial
and sustainability performance

Value drivers
 Analysis by different metrics: Past year
reports and data are used to generate
insights on future outlook by different
metrics e.g., revenue, growth, profits

 Scalable platform: Financial ESG reports

are most vital for assessing company
outlook. Insights from market trends,
product trends can be generated by
expanding

Page 184 Return back to meta data

 Use case category:

Sales agent - enables retailers to deliver persona-based, conversational

L4: Implemented with client
L3: Built for EY

journeys
L2: Built as a Prototype
L1: Still conceptual

13 Case example: Sales agent

Use Case
The client wishes to set up a personalised sales agent for enhancing the customer sales experience. It wants to provide customers with seamless persona-based
purchase journey by reducing the time to navigate the webpages, enabling virtual product trials, generating product comparisons and collecting feedback.

“AR Try On without Value drivers

visiting a store” “360-degree view of the customer”
 Improved user experience: Gen AI
enabled voice-based conversations in
multi-modal format with customers

 Personalised sales: Gen AI sales agent

enables product comparisons and virtual
product trials for persona-based customer
experience

Page 185 Return back to meta data

 Use case category:

Product recommendation - a digital assistant to support customers during

L4: Implemented with client
L3: Built for EY

product search and selection L2: Built as a Prototype

L1: Still conceptual

14 Case example: Product recommendation

Use Case
The client wishes to set up a product recommendation engine with AI enabled features for user friendly interface and enhanced customer experience by
reducing search time, generating product suggestions, seamless transaction and cross-category recommendations.

“Intuitive chatbot interface” “Easy conversion to transaction”

CLIENT Search Value drivers
CLIENT
 Improved user experience: Intuitive
chatbot interface generates key product
5%
recommendations based on customer
needs

 Seamless purchase experience: AI

enabled interface provide seamless
purchase experience by easy conversion
“Top 3 product suggestion matching “Cross-category product recommendations on to transaction
customer needs” request”
 Long term customer value: Based on
recent customer purchases, the tool can
generate insights on cross-category
product recommendations

Page 186 Return back to meta data

 Use case category:

ESG GenAI Benchmarking - generates company’s ESG benchmarking report

L4: Implemented with client
L3: Built for EY

based on annual sustainability report and open web information

L2: Built as a Prototype
L1: Still conceptual

15 Case example: ESG GenAI Benchmarking Tool

Use Case
The client wishes to set up a centralised tool for enabling different functions of the organisation to benchmark the ESG parameters for effective
implementation of ESG initiatives.

Value drivers
 Detailed benchmarking report: Provides
benchmarking against industry players
and outside industry players

 Enhanced ESG performance : Functions

across the organisation at various levels
can conduct benchmarking for effective
ESG target setting and implementation

Page 187 Return back to meta data

Generative AI has the potential to destroy business models and entire markets as
products and service can become obsolete, replaced or delivered by new competitors
X Case example: Business model and market impact assessment USE CASE | ILLUSTRATIVE

Business Model Risk Deep dive – AI Risk Assessment

Generative AI will become fully deployable and Inputs

accessible so quickly that profitability and even
Problem
entire business models may be at risk for some
companies, requiring swift adaptation. Client data Questionnaire Publicly available EY expertise/data
(quantitative / among business data (qualitative)
qualitative) partners
We run a comprehensive assessment of the risk
caused by generative AI considering three key
Solution GenAI risk assessment levers and model
dimensions: the type of risk, its typology of impact,
and its timing
Risk type Typology of impacts Risk timing
► Obsolescence ► Commercial ► Short term (<2 years)
Identify: Substitution ► Financial Medium term (2-5 years)
Realized ►
► Operational
►
► Obsolete Business models Competition/New entrants Long term (>5 years)
business ►
► Legal
►
► Outdated Operating models ►…
impact ► Disrupted industries

Generative AI scoring
How can GenAI afffect….
Business model
Strategy &
disruption Offering Operations Organi- Technology Market Business Perfor-
Innovation & Process zation & Model mance
Innovation

Solutions OpenAI

Source: Desk research; EY-Parthenon project expertise

Return back to meta data

GenAI use case
Oceania and APAC
Significant efficiency and cost savings through GenAI code conversion (1/2) Use case category:

Gen AI was leveraged to deliver code conversion resulting in significant

L4: Implemented with client
L3: Built for EY

savings (>85%) compared to the traditional methods for the customer L2: Built as a Prototype
L1: Proof of concept

Summary
Dimensions Traditional estimate Gen AI using GPT
 As part of the cloud migration initiative, the customer had a challenge
around conversion of legacy code (SQL and Dbt SQL) to BigQuery Volume of Code ~500 legacy GSQL and Dbt ~500 legacy GSQL and
(Please refer slide 3 for more details) converted SQL code Dbt SQL code
 EY team collaborated with the client team to understand the challenge
and applied GenAI in a quick POV for client showcase
 The results were outstanding with significant effort, time and cost Target code format GCP BigQuery GCP BigQuery
savings for the customer
 The accuracy was initially at ~80% and post 1 iteration of quick tuning, it ~100 FTEDays using
Code Conversion effort ~700 FTEDays
was improved to ~90% GPT
 The customer appreciated the EY solution and has requested EY to
support fast tracking of cloud migration
Execution Time required ~6 months ~1 month
 Moreover, EY is also currently requested to explore further
opportunities to fast track data migration and testing/reconciliation
using GenAI capabilities ~70% based on past
Accuracy >~90%
experience

Learnings Cost ~$850K - $1M ~$120K - ~$150K

 Withinthe broader solution, we need to be specific about where and

how GenAI can be leveraged to drive efficiencies Effort savings : >85%
 Adopting GenAI for code creation in this context accelerates the dev
Time savings : >80%
component, however testing and reconciliation also need strong Savings
Cost savings : >85%
automation practices to achieve full benefit

1.Generative AI projects (PoC or otherwise) completed with the client, at the client’s expense

Page 190 20 August 2024 Return back to meta data

Case study: Public Cloud Migration (Accelerated with Gen AI) (2/2) Use case category:

EY supported the design and delivery of migrating our clients core data
L4: Implemented with client
L3: Built for EY

platforms from legacy on-premise solutions to public cloud L2: Built as a Prototype
L1: Proof of concept

Business challenge
Key Deliverables
 Platform migrations are a main-stay of all organizations as both technology and client needs evolve
 Typically, clients have legacy systems, code, and data that have been built up over decades of investment, and this poses a 1. Cloud migration plan
significant challenge to migrate to modern technology, e.g. Cloud & AI Platforms 2. Generative AI code conversion
 Key challenges include managing the transition of existing services from current data platforms to new cloud platforms, accelerator
while also migrating & converting workloads from legacy-technology to new 3. Migration of data assets from
 This undertaking requires meticulous planning, coordination, and delivery management legacy platform to public cloud

Data Product Architecture Generative AI Code Migration

 Defined a data-product architecture with our  Developed a Generative AI Code conversion &
client, the transition from monolithic data test automation suite: This automatically Benefits to Customer
pipelines to modular, service oriented data converted code from one language to another.  Rapid transition of data services
products which can be maintained, extended, Our toolset included automation components from legacy platforms to public
and enhanced independently into the future. which integrated with client source-control to cloud
 Defined a data-mesh with our client to shift enable CI/CD type integration with code-bases  Reduction in technical and delivery
data usage away from solely being reporting to enable “continuous conversion” and risk from transitioning from legacy
What we did focussed, to being a scalable, flexible asset “continuous testing” of code conversion platform
which can be used in analytical, reporting, and  This achieved a significant uplift in speed of
operational use cases conversion and migration
(Please refer to Slide 3 for efficiency details)
Accelerators
Data Migration  Generative AI Code Conversion
accelerator
 Planned & Delivered the end-to-end migration of several data pipelines for our client, coordinating
activities from Design, Developing, and end-user testing, reconciliation, and quality acceptance
1.Generative AI projects (PoC or otherwise) completed with the client, at the client’s expense

Page 191 20 August 2024

 Use case category:

Using Generative AI to reduce a Big 4 Bank’s test data preparation time by

L4: Implemented with client
L3: Built for EY

over 70% L2: Built as a Prototype

L1: Proof of concept

The industry is witnessing an upsurge in adopting synthetic data due to challenges with labour-intensive manual test data
preparation, heightened reliance on data-driven decisions, rising privacy concerns, and limited access to real data in specific sectors

Challenges Impact Solution

Faced with the demand for faster Increases the time taken to release Leverage Synthetic Data Generation and
delivery and sustained quality, teams are robust and reliable data products into reduce the test data preparation time
exploring optimization opportunities production. by >70%

The testing data preparation

The test data preparation is Repeatable process for generating
impacts 30% of the Data
manual and labor-intensive Synthetic Data for any data product
Product Development Process.

Utilize metadata-driven strategy to

The existing testing procedures Disrupting delivery schedules
guide effective data anonymization,
lead to prolonged testing and hindering agile change
safeguarding privacy while maintaining
timelines. implementation.
data utility and integrity

1.Generative AI projects (PoC or otherwise) completed with the client, at the client’s expense

Page 192 20 August 2024 Return back to meta data

 Use case category:

Case Study: Gen AI use-case identification and prioritisation

L4: Implemented with client
L3: Built for EY

Achieving business uplift through technology for a large mining company L2: Built as a Prototype
L1: Proof of concept

The challenge EY’s DnA solution The benefit

A large mining company required external impartial expertise to
elicit key business value uplift within the service management
department. There was a requirement to ideate and prioritise use
cases based on both business value and technological feasibility
 EY’s role included stakeholder engagement and workshops
enabling the identification and prioritisation of 35 use cases. This
resulted in a Proof of Concept to be identified that aligned with Consistency Accelerated Implementation at
the company's technology platform OKRs (objectives and key across focus areas business value speed
results)

Stakeholders involved included business partners, internal  Provided initial suggested business-ready use-case candidates for The use cases generated helped substantiate:
customers and end-users. EY was engaged to provide independent refinement within workshops  Accelerated approvals for appropriate compute and staffing
analysis and prioritisation based on workshops and interviews  Defined and implemented a robust and rapid approach for use- resources
case identification and prioritization using agile and flexible  Accelerated sign-off from governance stakeholders across cyber,
delivery with stakeholder involvement in all steps architecture, legal and risk
The key challenges to overcome were:
 Worked with internal teams in adapting the approach for internal  A shared understanding of all key stakeholders of approval times
 Impartial assessment of perceived business value – clarifying and
ownership of use-case selection and requirements to unlock business value
resolving multiple conflicting preferences among stakeholder to
 Short-listed use-cases to de-risk and ensure the best likely  A technical and business-ready path forward for achieving a PoC
align on business impact
outcome designed to both capture business value and build the client’s
 Finding the “so what” – quantitively assessing feasibility based on
 Scoped and co-created project run-sheets and timelines around internal Generative AI capability
cost, time, and risk-management to determine how perceived
values aligned to OKRs selected use-cases and provided several options with varying risk
levels and timelines
 Agility to focus on evolving key areas of interest – while there
 Identified governance-level risks and determined the best
were extensive use cases identified, the stakeholders needed the
capability to adaptively re-prioritise based on evolving business technology ownership structure within the organisation
focus  Conducted 3rd party vendor interviews leveraging EY’s
relationships to ensure to potential future roadmap overlap
 Provided technical advice on integration, data maturity
considerations, platform suitability, solution architecture and
best-practices given existing vendor relationships
1.Generative AI projects (PoC or otherwise) completed with the client, at the client’s expense

Page 193 20 August 2024 Return back to meta data

 Use case category:

Responsible AI – Regulatory Landscape & Risk Assessment

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

The challenge EY’s DnA solution The benefit

 The client was looking to understand the global state of  EY compiled and presented a report on the Regulatory  Regulatory landscape report will help the client understand
Responsible AI regulations globally landscape across its operating markets including the both, the present state of AI as well as near and medium-
maturity levels and supporting policies term changes that will influence any AI initiatives that
 Being aware of how unregulated AI is at the moment
might be local or cross-border in nature
across the globe and requiring a sharp focus on current  The report provides the following details for global AI
markets and regulations around these, The client reached regulations:  With regulatory landscape reports available along with
out to EY to get an understanding of prevailing ethical AI EY’s AI framework, the client can work towards defining its
– AI Regulation trends across jurisdictions
principles, guidelines, and supporting regulations (such as own AI controls to place guardrails for ethical and
Data Privacy, Cyber Security) in the countries it operates – Guiding Principles to set the guardrails for any AI trustworthy AI models
to proactively embed guardrails in its operations while initiative
 Help the client to be proactive with the help of a
embarking on AI initiatives
– Supporting regulations (like data privacy regulation, and foundational risk assessment framework to build upon
 The Head of Data intends to have AI at the core of its data cyber security regulation) that need to be uplifted in an which can be embedded in their processes while embarking
transformation journey. The client was looking for a organization on an AI initiative involving any data considerations
Responsible AI framework to have trustable AI systems
– Infringement case studies with lessons learned  The foundational risk assessment framework will help the
developed
client develop a suitable and effective AI risk management
– Future trends in the AI regulatory space
strategy well suited to and aligned with their business
– List of Regulatory bodies for the jurisdiction or country vision. Once completed, they will be well equipped to
proactively manage AI-related risks by using the right
 Additionally, we provided a foundational Risk assessment
controls as needed
framework based on the principles from different
jurisdictions to evaluate the AI programs in consideration.
The framework was based on EY Trusted AI Framework
– A decision tree to classify the risk tiering of AI models
– An enriched risk assessment questionnaire combining
EY’s expertise and insights

1.Generative AI projects (PoC or otherwise) completed with the client, at the client’s expense

Page 194 20 August 2024 Return back to meta data

 Use case category:

PoisonGPT – Information from the Poisons Standard retrieved easier,

L4: Implemented with client
L3: Built for EY

faster and more reliable L2: Built as a Prototype

L1: Proof of concept

2023 | AUSTRALIA
The challenge Our Solution Key Outcomes
The Poisons Standard, is a 600-page regulatory document  EY teams created a chatbot solution, PoisonGPT, using  EY successfully delivered PoisonGPT tool, which met all the
that governs the classification and scheduling of medicines OpenAI’s gpt-4 API and embedding models with a client’s major business requirements
and poisons in Australia includes the classification of structured parsing over the Poisons Standard
 The PoC demonstrates EY’s capability in end-to-end
medicines and chemicals and the provisions about containers
 Extensive prompt engineering further enhanced the implementation of chatbots, leveraging Generative AI
and labels, and recommendations about other controls on
chatbot’s ability to produce tailored responses to the
medicines and chemicals
users’ queries
 A web app was designed using stream lit to create a simple
Key shortcomings of the Poisons Standard include: yet intuitive UI
1. It is difficult to navigate and find information  The solution also allows the user to upload their own
2. It is difficult to understand all the exceptions and rules document to extend PoisonGPT’s knowledge base
3. Potential for human error extracting information
4. Information about a single drug can be spread over
multiple sections

1.Generative AI Proof of Concepts (PoCs) that have been completed internally where EY is the client, or completed for the purposes of client pursuits

Page 195 20 August 2024 Return back to meta data

GenAI use case
EY Tech Consulting
USE CASE

Governance & Risk Mitigation

Organizations can de-risk and accelerate their development and deployment of capabilities by anchoring AI Governance in regulation,
Solution Statement principles, and strategy. Layers of regulation, policy, and technical requirements will shape how Generative AI is applied within the
organization

Regional Regulation

Client Values

Client Strategy

Client Policy

Tools and Architecture

Generative AI Mobilization

Business Model Use-Case Technology

Transformation Prioritization Strategy

External Client Technical

Page 197 20 August 2024 Return back to meta data
USE CASE

Generative Product Design Assistant

EY’s Generative Design Assistant unlocks limitless creativity by combining AI-driven consumer trends and insights along with pre-trained
Solution Statement custom models that revolutionize the design process. Reduce overall costs for prototyping and packaging and create offerings that are
in-line with current and emerging trends

1st Party Data

AI-Driven Consumer Insights AI Generated Prototype

Page 198 20 August 2024 Return back to meta data

USE CASE

Revenue Growth Management & AI

EY has developed a consumer commercial analytics platform powered by AI to quickly identify and recurrently tap into growth
Solution Statement opportunities across price, promo, assortment, and trade. With AI powered scenario engines that supercharge predictive and
prescriptive decision making to drive growth

Page 199 20 August 2024 Return back to meta data

USE CASE

Algorithmic Edge Advances Media-Mix Modeling & Scenario Planning

Traditional analytics process and visualize outputs. With Generative AI it’s possible to go beyond search and retrieval and introduce
Solution Statement
predictive scenario generation while also driving accessibility to marketing insights without being a data analyst

Page 200 20 August 2024 Return back to meta data

USE CASE

Supply Chain

EY is integrating AI across all phases of supply-chain. From Planning and procurement, production, distribution and logistics
Solution Statement
to in-store operations

Predictive maintenance
Augmented cognitive inventory
Procurement bots Track machine performance, predictive
management through AI-generated
Use of AI enabled bots to directly interact maintenance & resilient operations using AI
insights that eliminate under/excess Facial recognition technology
with business partners and vendors powered analytical tools
stock Use of AI technology to identify customer
related to procurement queries
and measure shopping frequency

Retail Supply Chain

Planning and procurement Production Distribution & logistics In-store operations

Plan distribution and order fulfilment

Autonomous shopping carts
requirements per AI driven demand analysis
Using AI enables shoppers to scan items in
Manufacturing Operations the cart, self checkout and pay via in cart
Utilize AI/ML to sense product quality, recognize card reader
inventory location and mine product attributes in Delivery route optimization
real-time on conveyor belt Using machine learning to create efficient delivery routes by
considering delivery locations, real-time traffic data and
driver schedules

Page 201 20 August 2024 Return back to meta data

GenAI Use Case
Insurance
Gen AI Driven FNOL

Return back to meta data

FNOL: Summarization and extraction: Solution

Business Problem Proposed Solution

Incomplete and lack of comprehensive information capture  GenerativeAI summarizes information from call transcriptions, call notes, and data
during FNOL results in substantial down stream challenges for captured in CSR systems into multiple cogent documents per call
an insurer-
 Generative AI is also used to extract entities and data from transcripts and call notes
 Timelimitations drive CSRs to not capture all needed
 The documents are indexed and packaged together per claim
information resulting in potential leakage
 CSRs and others in the claim handling team can find the document summaries per
 Richinformation shared by the customer in verbal
claim to find relevant information to support their claim handling needs
descriptions is not included in claims handling resulting in
erroneous handling of claims  CSRmanagers can find claim documents to assess the performance of CSRs and
create feedback to improve CSR performance
 Atthe end of the call, CSR review of captured data &
additional facts is laborious per call resulting in accumulation
of errors across multiple calls per claim Context Architecture
 Review of FNOL activity across CSRs by supervisors to
improve team performance is time consuming and results in
lack of continuous improvement of FNOL performance CALL
TRANSCRIPTS

CALL NOTES

FNOL SUMMARY & INDEX & DOCUMENT SEMANTIC

RECORD EXTRACT PACKAGE PACKAGES INDEX
EXPORT

Page 204 20 August 2024 Return back to meta data

Gen AI supported FNOL: Data Flow

1 Store audio 3 Get summary 5 Enrich 7 Personalized Communication

Store an audio recording Feed the transcript to Enrich the information Generate customized email
of a conversation between Generative AI with a with additional claims confirmation from FNOL
the claimant and the CSR prompt to summarize the center information on the transcript
key points from the text claimant

2 Create transcript 4 Submit 6 Sentiment & Tonality 6 Allow viewing

Create a transcript of the Submit the Generative AI Infer sentiments and Provide an interface
audio recording using a summary to Claims center emotions from FNOL for the adjuster to
speech-to-text model transcript view the Generative
AI summary

Page 205 20 August 2024 Return back to meta data

Gen AI Driven UW

Return back to meta data

Underwriting : Extraction & Knowledge Management using Q&A : Solution

Business Problem Proposed Solution

 Application
underwriting in Insurance involves components of  Use of Generative AI to
manual and time-consuming process. There is opportunity to
– Extract and store content from applications form, guideline documents and policies
use GenAI and add more efficiencies to help save on cost and
time. – Identify and present key entities from the application
 Some of the challenges faced are – Summarize information from various sources and present to the underwriter as text
– Less standardization since different insurance products – Provide an interactive QnA system which supports the underwriter to understand,
require different application forms and information from the assess risk and take decision on the application
applicant
– Large knowledge based on documents across past
Proposed Flow
applications and regulations has dependency on human
knowledge
– Extraction, storage and use of key information from the
forms is manual and not standardized across businesses
and applications
– Underwriter has to go through sections within large
application forms and past data to notice any potential red
flags. There is scope to use AI and reduce the time taken to
underwrite an application

Page 207 20 August 2024 Return back to meta data

Gen AI Supported Underwriting: Data Flow

1 Extract email 3 Get key entity Q&A system

5
Attachment information
Create a Q&A system to
Extract the attachment Feed the extracted text to answer user query based on
from the emails and Generative AI with a the historical forms and
categorize into form prompt to extract the key guideline docs by enabling
types and LOBs entities ChatGPT model

2 Text extraction 4 Generate summary 6 Allow interactions

Extraction of the text Feed the extracted text to User interface to view
from forms and guideline Generative AI with a the Generative AI
docs. Process the prompt to generate summary, key entities
extracted text to filter summary and access the Q&A
the sensitive information system
and further processing

Page 208 20 August 2024 Return back to meta data

Gen AI SAP EY use cases
SAP AI use case

EY Intelligent Transformation Platform

Accelerating the Move to SAP S/4 HANA with AI & RPA

Challenge
Automating processes, such as
 Complex and potentially costly migrations from ECC to SAP S/4 HANA
general ledger reconciliation, Average Business Value for Customers
goods-receipt and invoice-receipt  Minimal business case to drive and enact change and move to SAP S/4 HANA CALCULATION METHOD / ASSUMPTION IN NOTES
(GR/IR) clearing, and bank
transactions, based on decisions  Migrations that do not consider transformation led business case

** **
from SAP HANA PAL algorithms
and RPA Solution
 EY ITP for SAP S/4HANA is a solution that helps with the migration to SAP S/4HANA®, transformation migration time reduction
Automatic lift-and-shift of which is an essential step for clients who want to stay ahead of competitors. While the led business case (days vs. weeks or months)
configurations: customer vendor migration to the new SAP S/4HANA can be extremely time-consuming and daunting for
integration (CVI), GL migration and clients having highly customized systems, ITP comes to their rescue configuring and
automated approach to generation

** **
automating the processes to translate, transform and migrate data. It helps the clients
of UiPath scripts to also review the impact on security roles and process controls along with testing and
validating the migration from SAP ECC to SAP S/4HANA
high reduction in clean core focused including
Master and transaction data Benefits error rates automated code remediation
migration; process analytics scans
running business process in client  Significantly
reduces risk, complexity, cost and duration of an SAP S/4HANA migration
system or reimplementation through automation
 Intelligent Design:
Business case and migration type — The EY tool set informs a Customer / LoB Spotlight: *Customer Name*
Brownfield, Bluefield or Greenfield approach through analysis of the existing solution
client solution  Sysmex
Automation generation of analysis
 Intelligent Move:
Migration accelerators — The automation suite automates
for office reports; automatic ABAP  Knorr Bremse
custom code remediation configuration, data, custom code and simplification activities
 Lamb Weston
 Intelligent Deploy:
Deployment accelerators — This suite automates testing, training
and processing, providing a post deployment monitoring solution to track business  SSM Health
process changes  Discount Tire
EY CVI automate being an effective
mechanism to download or extract  Embeds new technologies, including RPA, AI, graph database technology and process  Fiserv
data from the client's SAP mining, into SAP-on-Azure and SAP public cloud as a central component of the
environment, covering the complete migration  Eldorado Gold
life cycle required for CVI in one  Multiple others
platform by completely automating
all the manual step
Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 210 Return back to meta data

SAP AI use case

EY Accounting AI for Payables

AI infused automation of payable processing

Challenge
 Highly manual and FTE intensive payables accounting
Automated invoice data extraction
Average Business Value for Customers
 Highcost of payables accounting (e.g., centralization, shared services and post-merger CALCULATION METHOD / ASSUMPTION IN NOTES
integration)

** **
 Excessive backlog in payables
 Quality issues and compliance issues in payables accounting
Prediction of posting (account, cost
center and tax key) on line-item
28% avg cost 50% FTE reduction
level, especially for non-PO invoice reduction
Solution
 EYAccounting AI for Payables is the next level of end-to-end (E2E) invoice processing

** **
automation solution. It uses ML, deep learning and anomaly detection AI techniques;
AI enabling PO matching on line-item and sets new standards in terms of efficiency, quality and sustainability of the
level purchase-to-pay process 5-20% reduction 5-20% reduction in error
in error rates rates

Benefits
 28% average cost reduction for clients of all sizes
Anomaly detection to identify Customer / LoB Spotlight: *Customer Name*
inconsistent or incorrect invoice  More than 50% reduction in FTEs needed for processing invoices
 Complete reduction of historical error rates (5%–20%)  Automotive Supplier (33 countries)
 Near-time processing, also for large invoice volumes  Medium-sized manufacturer (Germany, Sweden, UK)
 Sustainable automation rate, due to self-learning AI  Large non-food Retailer (Germany)
Prediction of approver for non-PO
invoices and automated workflow  Compete transparency on invoice processing efficiency  Health Care manufacturer of medicine and equipment (16
routing
 Complete transparency on internal cost and cost overruns caused by individual countries in Europe)
suppliers • Line-item level postings for all invoices (also non-purchase order (PO))
 +50 further clients in pipeline in GSA and USA
leading to complete transparency for procurement

Reporting and Analytics UI

Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 211 Return back to meta data

SAP AI use case

EY Intelligent Spend Analytics

AI infused spend optimization

Challenge
 Lack of visibility on total spend across Enterprise
Average Business Value for Customers
Gather spend data from multiple  Complex hybrid source systems CALCULATION METHOD / ASSUMPTION IN NOTES
SAP systems (and non SAP  Excessive spend/cost across Enterprise
systems)

** **
 Compliance risks & lack of forecasted data

Spend cost Compliance/Risk reduction

Solution reduction
 EY Intelligent Spend Analytics is designed to provide a 360-degree view of
organizational spends, by gathering relevant spend data from various SAP solutions
Combine into unified data mart with (like SAP ERP – ECC and SAP S/4HANA, SAP Ariba, Fieldglass, Concur and non-SAP
compliance checks data sources) and create an ‘Unified Data Mart’. Organizations have large, complex and
hybrid landscape solutions in-place to support their business models and use multiple
solutions for various types of spend categories. Existence of data in various cloud and
on-premise solutions makes it difficult to use for reporting. EY Intelligent Spend
Analytics will help fill in this gap by providing unified data, which can later be visualized
in an intuitive dashboard to derive insights into company’s spend and to identify saving
opportunities as well as reduce compliance risks
Customer / LoB Spotlight: *Customer Name*
Provide Analytics Dashboards
including predictive capability Benefits  Solution under development, not GA yet
 360 degree view of organizational spend
 Risks and compliance monitoring
 Plug-and-Play Solution to avoid IT costs, powered by SAP BTP and Intelligent
Technologies
 Monitor and control spends, Spend Prediction
Conversation AI (Chat-Bots) for
 Data-driven decision-making with AI capability
generating reports on-Demand
(planned for future releases)

Target Customers (of the solution the AI capability is embedded within)

 Cross Industry Applicability, Financial Services specific content provided as well

Page 212 Return back to meta data

SAP AI use case

EY CFIN SAP Error Reporting Dashboard in SAC

Accelerating the Move to SAP S/4 HANA with AI & RPA

Challenge
 Noconsolidated view of the errors relating to the documents posted in CFIN system for
a set of company codes
Average Business Value for Customers
CALCULATION METHOD / ASSUMPTION IN NOTES
Post documents in source SAP ECC  Ability to look at the errors for a particular time frame and for a set of company codes
system

Solution
**
transformation
led business case
**migration time reduction
(days vs. weeks or months)
 Central
Finance uses the side car approach to load/ re-post documents into CFIN
system to get accelerated benefits of S4 Hana system without disrupting existing

** **
SLT Replicates the documents into systems
CFIN Sytem  While doing so a lot of them (20%) goes in error for variety of reasons
 This
development uses NLP techniques to read the error text, do the pre-processing, high reduction in clean core focused including
tokenize it, put them on a vector space and cluster them error rates automated code remediation
 Post
clustering the keywords are extracted for each centroid and the clusters are
mapped to appropriate SAP labels for the keywords

Customer / LoB Spotlight: *Customer Name*

Document is either successfully
posted or in error in CFIN system.  Multiple others
Benefits
 Single point view for all the errors related to documents replication into CFIN system.
 Embeds new technologies, including RPA, AI, graph database technology and process
mining, into SAP-on-Azure and SAP public cloud as a central component of the
migration
If the document is in error, and
integration with SAC exists, We can
look at the errors in SAC and its
categorization with the help of AI.
Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 213 Return back to meta data

SAP AI use case

EY SOFIE
Smart Secure Solution For AI @ EY

Challenge
AI-driven solutions are leveraged to
 Business requires a multidisciplinary approach involving collaboration between
solve many critical business
technical teams, legal and compliance experts, user experience designers,
Average Business Value for Customers
problems. There are several threats CALCULATION METHOD / ASSUMPTION IN NOTES
that could threaten the management, and other stakeholders. It's important to strike a balance between
implementation process of these security, usability, performance, etc.

** **
solutions

transformation Enhance existing

Gather a dataset of clean images
that you'll use to train the
Solution led business case cybersecurity solution
autoencoder. Ensure the dataset is  EYSOFIE identifies the attacks present in the dataset used to train the machine
representative of the types of learning models that run on either the client environment or the SAP Business

** **
images your application will enc Technology Platform. The solution leverages the attack and intelligent capabilities of
GAN(Generative Adversarial Network), trained to defend against both known and
unknown attacks. Based on the insights provided, the necessary service actions can be high reduction in Helps to check Data Privacy
taken to improve the model performance. Industries get benefitted from SOFIE are error rates in AI models
Train autoencoder on your clean
dataset using the squared error loss
Retail, Healthcare, IT services, Government, Defence, etc.
function between the original image
and the reconstructed image
Customer / LoB Spotlight: *Customer Name*
Benefits
 Vulnerabilities of AI models and algorithms  Solution under development, not GA yet
Apply the adversarial attack
 Manipulation of data used in AI systems
algorithm to the encoded
representation of an image Then,  The exploitation of computing infrastructure used to power AI systems
decode
 Data poisoning
 Environment variations

 Credible and reliable feature identification

Evaluate the performance of your
adversarial trained autoencoder on
both clean and adversarial images
through its accuracy
Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 214 Return back to meta data

SAP AI use case

EY IWM
EY Intelligent Warehouse Management – Storage Area Determination

Challenge
 EY intelligent warehouse storage area determination is leveraging advanced
grouping and AI prediction algorithms to automate and improve warehouse layout
Average Business Value for Customers
CALCULATION METHOD / ASSUMPTION IN NOTES
Group products based on historical setup
orders and deliveries

Solution
**
Optimize storage
strategy
**Enhance existing warehouse
solution
 Itis an integrated solution with SAP S/4HANA that predicts products that need to be
stored together and allows warehouse managers to review suggested storage groups

** **
Assign Storage Area Indicators to and act by using historical ERP data and updating ERP product master data. It also
products based on generated uses prediction algorithms to automatically decide on storage areas for new stock-
groups keeping units or products coming into the ware
High reduction in AI- driven solution
error rates

Benefits
 Dashboard to review result of storage groping algorithm based on historical shipping Customer / LoB Spotlight: *Customer Name*
Assign new product to the group data
and set storage area indicator
 Setwarehouse storage area(s) for each storage group and update product master data  Solution under development, not GA yet
automatically
 Run several case analyses by selecting shipping data and adjusting grouping
parameters (max number of delivery per group / max number of SKU per group etc..)
 Activateautomated predictive AI storage area determination for new SKUs, reduce
master data maintenance tasks

Put away

Target Customers (of the solution the AI capability is embedded within)

 Cross Industry Applicability

Page 215 Return back to meta data

SAP AI use case

EY SPACES
EY Intelligent Insights

Challenge
EY Intelligent Insight leverages
 Stores lack in providing a personalized experience, recommendations, special discounts
Advanced Analytics and Artificial
for individuals and M-Experience which todays buyers are used to.
Average Business Value for Customers
Intelligence (AI) to convert data CALCULATION METHOD / ASSUMPTION IN NOTES
into useful insights along with  Retailers are
also seeing needs to provide curated user reviews, intelligent score,
Augmented Reality to create an
transparency and analytics to their instore customers

** **
unforgettable and invaluable
experience for the in-store
customers
Optimised customer Transparent competitive
Solution
experience price analysis
 One stop solution for insights driven strategies
A training pipeline (Review  Price comparison of the in-store items with platform providers

** **
Sentiment analysis) is created in  Intelligent score of each product after comparative analysis
SAP DI to train the NLP model on
 Rating and buying trends of the item
history data Augmented AI-driven sentiment analysis
 Intelligent curated reviews
reality experience
 Making the result of analytics immersive on the items itself through AR/VR

Benefits
A training pipeline (10SpecScore) is
 Readyreckoner for retail customers resulting in an increase in sales by reducing Customer / LoB Spotlight: *Customer Name*
created in SAP DI to train the
regression model on history data. It
dropout rates by 70%
can be accessed from ML Scenario  Better in-store buying experience  Solution under development, not GA yet
Manager  Building trust and transparent competitive with online counterparts

Sentiment analysis is done and

reviews are published using
Transformers

Target Customers (of the solution the AI capability is embedded within)

 Cross Industry Applicability

Page 216 Return back to meta data

SAP AI use case

EY Smart Onboarding
Efficient and experience-driven virtual onboarding mobile application

Challenge
HR sends an invitation email with  Lack of a structured onboarding process and lots of manual tasks
instructions to access EY Smart
Average Business Value for Customers
 Issues with navigating through virtual communication with information overload CALCULATION METHOD / ASSUMPTION IN NOTES
Onboarding
 Lack of support during onboarding, especially during IT setup

** **
 Lack of engagement and personalization during onboarding
Employee access mobile app and see
checklist with tasks, ability to upload Onboarding time Employee engagement
documents and information about reduction increase
important contacts Solution
 EYSmart Onboarding is an application that streamlines the virtual employee
onboarding process by providing real-time status checks for checklists and document
All uploaded documents are uploads. Contact information for hiring managers, team members, and onboarding
processed using OCR, saving time in buddies is also available, along with a chatbot for FAQs and advanced process
manual receipt and processing of automation. Automatic notifications keep relevant stakeholders informed of status
hard copies of documents changes based on roles, and the IT team is up-to-date on hardware requirements to
reduce delays. The result is a more efficient and experience-driven onboarding process

Employee can ask a Chatbot

questions about company, pending Customer / LoB Spotlight: *Customer Name*
tasks, information about manager Benefits
etc.
 HRteam can save time and cost by uploading necessary information on the application  Solution under development, not GA yet
rather than sending by post
 HRteam can save time in processing information of new hires by downloading the filled
Important stakeholders are notified
forms from the application rather than waiting for it to arrive by post
about status of tasks and for
approvals  Higher
retention rates leading to less expense in rehiring for same roles due to a better
onboarding experience
 Simpler and better structured communication channel with all new hires

Informed and happy new hire

Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 217 Return back to meta data

SAP AI use case

EY Intelligent Contract Generator

Auto-Create Intelligent Risk Mitigated Contracts Leveraging Generative AI

Challenge
 Lackof ability to generate contracts to mitigate dynamic geopolitical risks across
enterprises
Average Business Value for Customers
CALCULATION METHOD / ASSUMPTION IN NOTES
 Increased contract management cost requiring intelligent automation
Generate contract which

** **
automatically mitigates the dynamic  Generic contract templates not meeting specific contract requests
geo-political risks

Low risk in contract Huge efficiency gain by

Solution execution (expect automating contract
 Orchestration algorithm designed and built using SAP BTP services 30% risk reduction) creation (days vs minutes) –
 Enterprise-tailored languagemodel using existing contracts of the enterprise and Approx 40K contracts are
subsequently utilizing the same to effortlessly produce risk-free contracts created on average for
 Leveraging Generative AI capabilities Fortune 1000

Creates contract which are specific

to its contract request data
Benefits
 Significantly
reduces risks in contracts for the enterprise since the clauses are more
relevant to the specific contract request mitigating the evolving geo-political risks
Customer / LoB Spotlight: *Customer Name*
affecting the contract
 Multiple others
 Hugecost and effort saving by automating the large volume of contract documents
generation in near-time
 Plug-N-Play solution built on SAP BTP reducing IT costs
 Data driven intelligent contract authoring uncovering hidden risks for the buyer
 Continuous improved automation due to self-learning algorithms
Automation of large volume of
 Embeds Generative AI technology supporting further innovations
contracts intelligently

Target Customers (of the solution the AI capability is embedded within)

 Industry agnostic solution applicable to all enterprises

Page 218 Return back to meta data

SAP AI use case

EY BTC iRPA
Automated A1 and PWD submission

Challenge
IRPA bots works in backend on
 Thetravel documentation needs to be filled manually by the users or the case worker
dedicated VDIs . From technical
team
Average Business Value for Customers
point of view – all bots are working CALCULATION METHOD / ASSUMPTION IN NOTES
in unattended mode (which means  The time and cost efficiency can be improved by implementation of the automatization
bot processing is triggered as soon

** **
as job is available and there is
available machine that bot can use).
However, from practical point of
view – bots are divided into two
Low risk in contract Huge efficiency gain by
categories
Solution execution (expect automating contract
30% risk reduction) creation (days vs minutes) –
 RPA Bots for automated A1 and PWD
Approx 40K contracts are
Attended Bots
created on average for
Used whenever new bot is deployed
or there was a change in existing bot
Fortune 1000

Benefits
Unattended Bots  Time and cost efficiency by proces automatization in:
Unattended bot runs on the Filling an online form – most frequently
dedicated VDI that is always online.
–
Customer / LoB Spotlight: *Customer Name*
Once the job is available in the – Filling Word/Excel document
queue – bot is automatically – Filling PDF (if possible)  Multiple others
executing that jo
 Submission of PWD/A1 form is done via:
– Online form submission – most frequently
– EYI (Interact) upload
Plug-n-play solution – Email submission to authorities or local (EY) country office

Generative AI capability built-in to

mitigate new risks
Target Customers (of the solution the AI capability is embedded within)
 Cross Industry Applicability

Page 219 Return back to meta data

SAP AI use case

EY Intelligent HelEYos
Training existing ML algorithm using genAI

Challenge
Usability: The platform caters the  Performing theend-to-end drone flight process, data ingestion and analytics solution
end-to-end flow starting from the through an integrated platform
Average Business Value for Customers
creation of drone flight request in CALCULATION METHOD / ASSUMPTION IN NOTES
SAP FSM to a seamless process  Enabling remote site intelligence through the analytical reports that have been derived
workflow of the request in HelEYos from aerial photos captured by drones

Regulatory and Suitability: Ensures

 Pro-active flagging of potential concerns seen in geospatial data
**
Faster, safer and
more efficient
**Improves data accuracy
compared to the traditional
compliance with regulatory
Solution
standards and suitability for drone method
operators, equipment and pilot  Leveraging Generative AI capabilities

** **
before the drone flight  EY HelEYos enables data processing analytics with geospatial visualization to automate
business processes. The platform is integrated with SAP Field Service Management
(FSM) for plant maintenance, that caters the flight requests from FSM to HelEYos and Real-time data for High operational
Processability: The acquired drone
integrates with SAP S/4HANA for work order creation with respect to the plant an improved efficiency
maintenance. The solution also offers maintainability of site equipment with its remote
imagery data will then be processed decision making
to obtain geospatial information site intelligence for stock verification drone use case
through photogrammetry
Customer / LoB Spotlight: *Customer Name*
Benefits
 Faster, safer and more efficient performance of remote asset inspections
 Power & Utilities, Manufacturing, Energy, Oil & Gas,
Actionability: Integrated to SAP Mining, Telecom, Engineering, Construction and
S/4HANA, maintenance order  The solution improves data accuracy compared to the traditional method
requests can be done in the form of
Operations
 Leveragingthe ease of drone image captures and real-time data for an improved
annotations from HelEYos
decision making
 Achieveoperational efficiency through streamlined processes as the solution runs on
an integrated platform
Safety: Tracks drone maintenance
records and pilot credentials
Surface inspection: Dust detection Target Customers (of the solution the AI capability is embedded within)
(RGB images)
 Utilities, Engineering, Construction, Energy, Oil & Gas, Manufacturing, Telecom, Mining
companies

Page 220 Return back to meta data

SAP AI use case

EY PUIC – AI DEBT Prediction

AI infused Debt prediction monitor

Challenge
 The unavailability of any monitoring process in S4 to evaluate and predict the paying
pattern of the customer,
Average Business Value for Customers
CALCULATION METHOD / ASSUMPTION IN NOTES
CDS views in S4 for customer data  Lots of cases of non-compliance in payment trend is notices in the Utility industry
across globe
 Need for holistic view of customer and its credit score
**
Better understanding
of Debt Trends
**In-Built Debt prediction
Solution
 PUIC offers ready-to-use management information dashboards and effectively

**
Consuming the different CDS in manages back- and front-office operations by leveraging SAP Analytics Cloud for real-
SAC time data analysis. It augments this data through EY processes and operational
experience, such as the use of Machine Learning (ML) to provide debt prediction Detailed overview of
capabilities
Customer insights

Benefits
 Increase transparency and traceability around Meter-to-Cash value chain Customer / LoB Spotlight: *Customer Name*
Configuring algorithm model in SAC  Provide a real-time snapshot of operational performance metrics
over these CDS  Asset Management
 Provide an always up-to-date answer to key questions such as, overview of number of
customer accounts, expected revenue, the debt they are generating and the risk profile  Finance
of said debt  Sales and Service
 Provide geographical representation of areas with the most business exceptions,  Metering,
revenue, debt and/or service requests to aid planning
 Plant Operations & Maintenance
 Measure bill accuracy, with breakdown of generated, delayed, reversed and adjusted
bills
Building customized analytical
dashboards using the S4 data

Target Customers (of the solution the AI capability is embedded within)

 Potentially every utility running IS-U could be a candidate for PUIC

Page 221 Return back to meta data

SAP AI use case

EY Power and Utility – Virtual agent

AI infused with virtual agent chatbot for customer

Challenge
A virtual agent's comprehension and capacity to respond to client enquiries are limited
as a result of supervised learning
Average Business Value for Customers
Generative ai to understand the CALCULATION METHOD / ASSUMPTION IN NOTES
intention of user query  Rely on pre-programmed responses and are unable to learn from user interactions and
clarity on customer commands
 Classification rely on trained model
**
robust
understating
**better response time
Generative ai helps the virtual agent
to Route the query to respective use
case Solution
A virtual agent assists customers with questions about invoicing, the creation of
tickets, filing complaints, and general inquiries. Given that it uses a pretrained model,
generative AI aids the virtual assistant in better understanding the user's question.
Additionally, the classification of ticket categories will be more accurate. It gives a
**
Better
classification of
**use case routing
Collect user data and call the api
more conversational approach to existing chatbot ticket category
from connected SAP system(C4C,
S4 ,FSM)

Customer / LoB Spotlight: *Customer Name*

Benefits  Utilities
 Unclear customer commands can be better understood and more robustly handled
Generate ai format the response in
conversational experience A more conversational approach can be taken, as the current chatbot is very stiff and
structured in the order it deals with commands
 Complaints/tickets can be automatically classified into the right category, making
resolution easier

Present the repose to user in

expected format which includes PDF
invoice
Target Customers (of the solution the AI capability is embedded within)
 Power and Utility Industry Sector

Page 222 Return back to meta data

EY Generative AI
Credentials Pack

Edition 3 – April 2024

Section Generative
divider over two
AI:lines or
Consumer
three lines

Page 224
 Use case category:

Generative AI Enabled Product Design

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

EY is building a proof of concept (POC) to help the client’s watch designers be more The solution streamlines design process and identify areas in the market
creative in the design of new products. For example, no longer will designers be limited to for watches that might become popular. Generative AI model is trained to
what they can visualize in their own mind. respond to prompts like, “design a watch inspired by a Ferrari”, or “make
this watch look more high-end”, and the model will create images of new
product ideas that meet those requirements.

Approach Taken (Phases and Key Activities) Value Delivered

 Concept visualization: Create realistic, three-dimensional renderings of
new product concepts
 Design optimization: Analyze the design data and suggest modifications,
creating better and more cost-effective products
Fine-tune Generative Work with designers Implement the White space
AI models on data to to further refine the solution with data identification, 3D  Material selection: Choose the best materials based on cost and
evaluate the image model and make the integration and a modeling etc. performance specifications
generation capability images designer-ready front-end interface  Prototype acceleration: Render accurate digital prototypes of proposed
product designs

Technologies  Concept evaluation: Perform “what if” scenarios and enhance the
process of comparing and assessing product concepts
 Stable Diffusion XL (SDXL)
 Google AI
– PaLM 2
– Google storage bucket
– Vertex AI

Return back to meta data

 Use case category:

Generative AI Factory and High Value Use Cases

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The Client had shown keen interest in collaborating with a partner on their journey of Generative AI adoption with a primary objective of substantiating the value proposition of Generative AI, through prioritized
proof-of-concepts, thereby laying the groundwork for establishing a Generative AI factory in 2024.

Solution Provided Value Delivered

 Reporting Automation:  Reporting Automation:
– Deployment of an automated reporting solution leveraging – ~50% increase in speed while developing performance reports
– Generative AI that can be replicated by non-technical users and its extensibility to copilot – Accurate execution of recurring analysis
 Product Listing Writer:  Product Listing Writer:
– LLM solution to generate product listings with client’s consumer reviews data using EY PLW – Automated process that allows simultaneous customization of multiple product listing copies
accelerator and evaluate the impact on existing workflow (process efficiencies) and the quality – Better personalization by retailer, season, events etc. driving e-commerce sales
of optimized product listings  Hot Care Agent Assistant:
 Hot Care Agent Assistant:
– Improved consumer experience due to faster and more accurate responses
– Empower agents with a responsive tool that promptly suggests solutions upon entering a
– Potential reduction in agent capacity
customer's issue or question using knowledge base retrieval and generative AI

Approach Taken (Phases and Key Activities)

Workshop and use case MVP execution and value MVP complete, Governance and Productionize MVP and Initiate
identification measurement 2024 Factory Plans Finalized Generative AI factory deployment
 Interactivesession to explore  Finalizehigh value use cases for MVP and  Complete MVP by incorporating UAT feedbacks  Define future state vision of the MVPs and deploy high
Generative AI capabilities and measure value through UATs  Develop a high-level Generative AI Factory value use cases
identify practical use cases.  Develop high level recommendations & implementation roadmap to build the future  Develop a Generative AI factory including demand
blueprints for AI governance, Operating model, state focusing on quick wins and long-term management, AI governance, Operating model,
architecture and Demand management opportunities Enterprise arch and development of use cases using
Generative AI

Return back to meta data

 Use case category:

Generative AI Center of Excellence and Use Cases

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client requested support in developing a Generative AI Center of Excellence in order to define a Gen AI operating model and design technology and data architecture to
execute Generative AI use cases at scale. In addition, the client was interested in the build out of 7 proofs of concepts to begin the execution of their Generative AI Strategy.

Solution Provided
EY worked with the client to understand their current operating model, technology architecture, and AI governance processes and documentation to conduct a gap analysis
against the future state vision. Based on that, the team created a Center of Excellence framework and roadmap for the client to realize their transformation utilizing
Generative AI. In parallel, a team of developers worked with the client to identify and implement 7 Generative AI PoCs.

Center of Excellence Framework and Roadmap 7 Use-case Accelerators Implemented

1. Design, Structure and Operating Model Contract Co-Pilot Reporting Co-Pilot
2. Innovation, Process and Value Measurement
3. People and Change Management Product Catalogue
Digital Recruiter HR Policy Navigator
4. Solution Architecture and Technology Creation
5. LLM Ops Framework
Product Improvement
6. Gen AI Risk & Governance Sales Co-Pilot
Analyzer

Approach Taken (Phases and Key Activities) Value Delivered

EY worked in 2 parallel workstreams. One which focused on the Center of Excellence  Accelerated PoCs: Client was able to experience the implementation of
Roadmap and Framework, the other supported the build of the 7 use-cases. generative AI at speed in their own environment
 Future Vision and Roadmap: Client is able to envision and implement a future
transformation utilizing Generative AI

Return back to meta data

 Use case category:

Store Ops Assistant

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

The client and its employees face challenges in understanding and responding to store EY worked with the client to build a store ops assistant that is a streamlit based web
associate queries on store operating procedures, such as large time consumption on app hosted in the Azure Cloud regarding store operating policies and procedures. It
response to repeated inquiries, less time for customer interaction impacting on sales, harnesses the ability of Generative AI to empower store operators to do their jobs. The
replenishment and shrink, impact on store associate confidence resulting in high OpenAI based chatbot allows store operators to answer any questions they might have
attrition and training costs, etc. regarding internal store policies quickly and accurately.

Functionality Flow
Approach Taken (Phases and Key Activities)
1. A working prototype of Store Ops Assistant application on a desktop
2. Development will be done in Dollar Tree’s Azure environment
3. UI will be built using Power Virtual Agents
4. User can change data, edit UI
5. All the data requested to be in Azure environment, Integration with
legacy systems will be in the next phase

Value Delivered
 Reduced store associate hours spent searching and fewer management hours responding to repeated inquiries
 Increased time for customer interaction, leading to potentially higher sales, replenishment and decreased shrink
 Store associate confidence resulting in lower attrition and reduced training costs
 Augmented and continually refined knowledge database
 Insights into store associates’ top issues

Return back to meta data

 Use case category:

Generative AI Chatbot POC

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

Chick-Fil-A requires help to efficiently manage their vast operations Two separate Gen-AI powered Chatbots will be designed and implemented. They will provide
ranging from point-of-sale modernization to advanced kitchen instant and contextualized responses to a variety of IT and process-related questions. This will
processes. At the heart of this challenge is the need to provide reduce time taken to identify & troubleshoot problems, as well as reduce inefficacies related to
immediate and accurate answers to various queries from Operators, inaccurate assignment or understanding of cases, and reduce the need for live human support.
Team Members, and Staff, while eliminating inefficiencies, reducing Use Cases:
calls, and reducing mistakes in the assignment of cases.
 Chatbot can analyze and answer IT and Restaurant-process specific queries
– Give users the ability to interact with back-end systems using natural language
 Chatbot can identify the correct forms and fill them in using information from requestor
– Ability to pull and interpret data from and push forms into ServiceNow

Approach Taken (Phases and Key Activities) Value Delivered

► Llama 7B, integrated with Bot Framework and AWS, powered the first bot in a 10-week  Proof-of-Concept: Prove as a foundation of Gen AI capabilities, promising
sprint long-term value via additional solutions (ex: Chick-Fil-A internal ChatGPT
► Employing 'prompt sharing’: this approach optimizes our compact LLM for complex implementation)
tasks, ensuring consistent execution
► Developed a bot using Power Virtual Agents/Copilot Studio to test the limits of the  Streamlining operations:
platform – Improving accuracy of responses
– Reduce time to respond to queries
– Reduce time to close cases
Technologies
– Reduce number of/time on calls with agent
► Llama 2 (7B) (Meta AI) ► AWS – Improving the experience of Chick-Fil-A staff and operators
► Copilot Studio/PVA ► ServiceNow – Improve user experience in finding information
► Bot Framework

Return back to meta data

 Use case category:

UX/UI Development for Custom GPT Solution

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client had developed and launched release 1 of their own version of GPT, an internal AI tool, in late 2023. In updating the functionality of the client’s GPT version for
release 2, the client would like to place greater emphasis on the user experience to make the tool more intuitive and drive adoption.

Solution Provided Approach Taken (Phases and Key Activities)

The team is proposing a custom interface that leverages many capabilities of GPTs  Week 1: Research and Analysis: Analyze the existing system and conduct
beyond just the chat functionality. Included is a left rail to choose between GPT models, interviews with key stakeholders to understand user sentiment about the
see chat history, and plugins like DALL-E. The center panel will house the standard GPT existing system and preferences for a future solution
chat interface, and the right rail will include dynamic functionality like filters and facets
that change based on conversation context, file browsing, suggested prompts, and file
 Week 2: Low Fidelity Mock-ups: Rapidly create schematic mock-ups of
viewers.
potential future-state solutions

Value Delivered  Week 3: Branding, High Fidelity, and User Interviews: Create design files
 The team is set to deliver high-fidelity screen designs for the client’s own GPT Release ready for hand-off to the client technology team for implementation
2.0 in early February to be used in development

 Week 4: Prototyping: Create clickable prototypes demonstrating key

 With this solution our goal is to help client users understand the many ways in which functionality to be used for development and in testing
their version of GPT can support their work and make their day to day more efficient

Return back to meta data

Section Gen
divider
AI:over two lines or
AM&M
three lines

Page 231
 Use case category:

LLM for Digitization of Rules of the Road

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution

Automated Driving Systems (ADS) must comply with traffic rules and The LLM for digitization of on-road rules harnesses the power of Generative AI for
regulations (“rules of the road”). Rules of the road are heavily fragmented, extracting and formatting all relevant rules of the road within a given geographic
not harmonized across different jurisdictions, not machine-readable, and open jurisdiction for a given driving scenario into the client’s database. As of now, the end user
for interpretation. must ask the question about rules on the roads, but in future updates the ADS will
automatically ask about the relevant rules based on the users location.

Approach Taken (Phases and Key Activities) Technologies

► Microsoft Azure
Collect Scrape traffic rules and Query Ask in natural language Extract Collect relevant sources
► Langchain
regulations from different about the relevant rules then identify and format
sources and jurisdictions of the road to search the requested rules

Output provides
expected content and
follows required Value Delivered
structure
► Improved efficiency: Streamlines rule
scouting process and saves time
Query provides extracting rules
task instructions
► Consistency and accuracy: Minimizes
errors and rule misinterpretations

LLM reads query ► Capacity building: Enhances ability to

and parses text scout rules for more jurisdictions for
Server stores
scraped data for answers different driving scenarios
Search ranks most Combined text is used
relevant excerpts as context

Return back to meta data

 Use case category:

Generative AI - Enterprise Strategy & Pilots

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A large automotive client was interested in understanding how they should set up the ideal Generative AI architecture, how they should govern Generative AI, and what use
cases they should be considering to increase employee efficiency and improve overall work quality and performance of their internal operations.

Service / Solution Provided

Marketing PoC Client Operations PoC Everyday AI Status – MSFT 365 Copilot
Leveraging Generative AI capabilities to automate the Leveraging Generative AI capabilities to automate the Leveraging Generative AI capabilities to power team
process of generating landing page content that is process of writing a call summary that is currently member productivity by accelerating day-to-day tasks
currently being executed manually by marketers and being executed manually by agents can play a crucial and improving overall work quality
agencies can play a crucial role in improving role in improving consistency and standardization
consistency and standardization

Approach Value Delivered / Challenges

1. Vendor Landscape Analysis - Explore how client can most effectively leverage Generative AI ► Improved efficiency: optimizing SEO, improving customer
capabilities coming to their existing platforms and applications experiences, and allowing marketers to spend less time
2. Use Case Assessment - Create a framework to assess use cases submitted to CAI leadership drafting content and focus on creating value and efficiency
and design a plan to pilot and test select use cases including Content Creation for Marketing,
and Client Operations ► Consistency and accuracy: ultimately improving reporting,
3. “Everyday AI” - Assist with the roll-out and change management for Microsoft’s Copilot tools actionable insights, and allowing agents to spend more time
and determine the most beneficial use cases and groups to pilot with with clients

Return back to meta data

 Use case category:

Generative AI Factory, Sales Forecasting Tool, and Chatbot

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client needed support to begin their Generative AI journey and prove quick value. EY supported them in deploying 2 AI use cases and establishing the foundation for an
AI Factory to support scaling of AI across the enterprise. The MVPs developed include a sales forecasting tool and a technician chatbot.

Approach Taken (Phases and Key Activities)

While working on the two use cases, EY will bring or AI  help business leaders rethink strategy and operations — from new business models and
playbook to life and leverage our EY.ai platform and Strategy products/services to redefined business process — by putting human and machine together
foundations to set the foundations to scale: at the center with a focus on creating value and managing risk.
 face transformation across the functions — supply chain, risk, finance,
Functional  customer experience — which will require reimagining processes and systems. EY’s
Confidence Transformation approach reflects a deep understanding of Komatsu’s performance indicators, dealer and
in AI supplier network, customers, and employees.
Functional Building the Confidence in  build user confidence in AI solutions to drive adoption and navigate a dynamic legal and
Transformation foundation AI regulatory landscape as AI continues to evolve globally, while establishing guardrails and
curated sources of truth to remain trusted.
 accelerate the adoption and integration of AI across the enterprise to support existing
Building the
technologies, capabilities, products and services by leveraging the EY.ai platform, including
Equipping foundation
service architectures, modules, libraries, data readiness frameworks, and MLOps processes.
EY AI and upskilling
Strategy Equipping and unlock the full potential of the workforce, including deploying new learning and upskilling
Playbook the 
workforce upskilling the programs, and providing documentation and training on new AI-enabled solutions and
workforce workflows.

Solution Provided Value Delivered

Set the foundation for and grow an AI factory with a portfolio approach to building Client quickly scaled up their factory by leveraging EY.ai processes, tools and
out applications with renewable processes, models and tools leading architectural blueprints

Return back to meta data

Section Gen
divider
AI:over
FSOtwo lines or
three lines

Page 235
 Use case category:

Call Summarization and Insights Capture from Advisor calls

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A large Wealth Management firm is working to foundational contact center capabilities using Generative AI to generate business insight and summaries of advisor / client
calls.

Solution Provided

Advisor Call Summaries

Develop 8 prioritized business insights to quantify  High level of accuracy of 8 prioritized business insights
completeness and accuracy and drive summaries generated  Summary Leveraging Business Insights

Approach Taken (Phases and Key Activities) Value Delivered

 Diagnostic Testing & Performance Improvement  Trained LLM model ecosystem with instruction dataset for Fine-Tuning and
– Testing framework enhancement Prompt Library
– Chunking strategy redesign
– Solution design for performance improvement  Re-Usable modules for future use cases
– Diagnostic Testing Framework
 Prompt Engineering – Iterative Performance Improvement Framework
– Zero Shot / Few Shot, Directional Stimulus, Knowledge Based – Approach for quantifying business impact
– Operational Performance Framework
 Model Fine-tuning
– Llama2, Falcon, GPT-NeoX and Mistral

Return back to meta data

 Use case category:

Global Consumer Bank leveraging Generative AI to Conform to

L4: Implemented with client
L3: Built for EY

Regulatory Feedback
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A large international bank received regulator feedback related to the timeliness controls testing as well as oversight and aggregation of control effectiveness assessments.
As part of their response the bank needed to identify and prioritize the population of key controls that would require remediation and then execute testing in a timely and
consistent manner.

Solution Provided
 Provide insights into the existing control environment through deployment of EY’s natural language processing tool to analyze the existing control inventories and
determine control description quality and ability to be tested
 EY was able to assist in this inventorying process and then assess the population of controls to determine which controls would be suitable to for testing and which control
descriptions required enhancements prior to testing

Approach Taken (Phases and Key Activities) Value Delivered

 Perform a 5W Analysis to better understand the key elements of a control as part of the  Lower effort & higher productivity: Identified 34% of controls that were
control quality assessment subject to enhancement which led to efficiencies gained in the testing
– The tool employs NLP semantic role labeling methodology effort
– The tool aims to identify and assign labels (e.g., verb, subject, etc.) to components of  High quality & consistent content: Provided analysis on the enterprise
a control description control inventory by line of business, control type, and quality
– Components with labels associated with the 5W category (Who, What, When, Why,  Better service & client engagement: Delivered a control testing
Where) will be bucketed prioritization plan consisting of approximately 2,000 controls
– A weighted quality score is assessed for each control

– Additional metadata are considered, including Inherent Risk Rating, Product

Criticality, and Business Unit as key prioritization factors

Return back to meta data

 Use case category:

Financial Services HR Department – ChatGPT for Enterprises with

L4: Implemented with client
L3: Built for EY

Guardrails L2: Built as a Prototype

L1: Proof of concept

Problem Statement
Implementation advisory of a ChatGPT-based chatbot solution integrated with Microsoft Teams for the purpose of creating and delivering training and learning materials to
its employees. The client wanted ChatGPT to be secure with guardrails and compliant with enterprise application guidelines Any open-end access could create risk for
enterprise.

Approach Taken (Phases and Key Activities) Solution Provided

 Client wanted selected people/dept’s in within the organization to access ChatGPT to generate non- EY Team helped to create a Secured Architecture for
proprietary content in secure manner, via Teams interface. proposed solution including:
 For additional security compliance client wanted all sessions to be tracked, logged and tracked for
compliance.  Teams interface integration
 Content Moderation, hallucination control and filtration were other requirements
 Context management via Azure Bot service
 Call logging, AD Integration, RBAC permissions
 Prompt engineering for reducing hallucinations & prompt
injection protection
 Performance and scalability architecture guidance
 Content moderation and filtering configuration via Azure
Content monitoring

Value Delivered
Solution provided secure and compliant way for the client’s
HR Training team to access ChatGPT for business purpose,
while being compliant with enterprise requirements.

Return back to meta data

 Use case category:

Generative AI Strategy & Roadmap for a Global Insurance Company

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Runoff insurers have a unique problem of managing books of claims acquired from other carriers for various products. This presents opportunities to gain efficiencies in
their operations by adopting AI/Generative AI in their business processes. Client requested EY’s assistance in developing foundational frameworks that would enable them
operationalize Generative AI and augment their business processes.

Solution Provided Approach Taken

 Defined a target state reference capability framework for ML and Generative AI
solutions
 Developed a high-level decision tree framework for solution component selection
 Developed starter architecture based on a choice use case solution pattern Target State
 Developed guiding principles and reference framework for getting started with Reference
AI Ops Architecture
 Provided recommendations for enhancements to address gaps in the Client AI
risk and governance framework

Value Delivered
 Actionable roadmap to establish foundational capabilities for AI adoption in
their business processes
 Ability to incrementally enable capabilities through targeted use case and with
Reference
dedicated Gen AI labs LLM Ops
Framework
 Ability to establish experimental environment through a reference architecture
framework and get started on use case build
 Ability to establish robust AI governance framework with
recommended enhancements/mechanisms to address heightened risks

Return back to meta data

 Use case category:

OpenAI GPT4 based Solution Financial Advisory

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A large Wealth Management firm is executing a program to integrate OpenAI into core platforms with the intent of (i) improving access to internal content by employees
(Financial Advisors, Service Associates), (ii) enhance employee productivity, and (iii) improve client experience.

Value Delivered
 Lower effort & higher productivity:  Better service & client engagement:
Accessing/Searching for information, tagging content, etc. Understanding user needs and experiences (sentiment, etc.)
 High quality & consistent content:  Reusable Assets for LLM Adoption and Scaling:
New content generated e.g., knowledge - title, keywords, FAQs, notes, etc. Testing/ evaluation suite and enhanced LRC governance, etc.

Approach Taken (Phases and Key Activities) Use Cases

 Rigorously evaluate vendors for contextual search, metadata generation, and virtual Developed Ongoing
assistant capabilities Internal Knowledge Management Call Center Analytics
► Virtual Assistant
 Prioritize tools aligning seamlessly with the organization's knowledge management ► Call Center to Client
► Search
goals ► Financial Advisor (FA) to client
► Metadata (Summary, Taxonomy,
 Seamlessly integrate internally developed tools with external solutions FAQs)
 Create an efficient knowledge management ecosystem
 Ensure smooth integration of selected tools into the existing operating model EY Support Areas
 Implement strategies for widespread adoption, especially among Financial Advisors
 Implement robust contextual search for quick and precise information retrieval Use Case Model Future MRM/Le Overall
Actively use metadata generation techniques for efficient document categorization Design & Testing State gal/Risk
 Program
Develop (Testing BAU Documen Support
 Deploy virtual assistant capabilities for intelligent and context-aware support, ment Suite) Design tation
improving information accessibility and

Return back to meta data

 Use case category:

AI Governance Framework and Target Operating Model

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Supporting a large financial service company with the design and implementation of its Artificial Intelligence (AI) governance framework and target operating model
including consideration for Generative AI.

Solution Provided Value Delivered

 Organizational Policy and Procedure Review and Update  Risk Management: AI / Generative AI framework help identify & manage risk
 AI Operating Model and Readiness Assessment  Quality & Performance: ongoing monitoring and evaluation of AI / Generative AI systems in Function

 AI Vendor Selection and Cyber Threat Scenario Definition specific procedures

 Data Management: AI operating model helps the supervision of data quality, data privacy and
 Communication and Training Framework
appropriate use of data

Approach Taken (Phases and Key Activities)

 Developed AI /Generative AI governance framework (AI Council and Working Group) and enterprise AI / Generative AI Governance Policy
 Reviewed Function specific policies and procedures including recommendations for AI / Generative AI-specific enhancements and corresponding policy/procedure language

 Design of target state AI / Generative AI operating model including end-to-end AI delivery life-cycle and corresponding responsibilities for teams as well as illustration of
operating model through client prioritized use cases and solution architecture
 Assessment of AI / Generative AI readiness comparing current state against target state operating model including observations, recommendations, capability maturity
matrix and roadmap providing a phased approach to arrive at the target state
 Develop vendor tool selection criteria that can be used to assess AI / Generative AI governance and monitoring and cybersecurity threat solutions
 Identification of potential cybersecurity threat scenarios as it relates to the use of AI / Generative AI
 Develop actions to roll out AI / Generative AI governance policy and operating model and design corresponding communication and training framework tailored to
different stakeholders (e.g., executives, business, and operational teams)

Return back to meta data

 Use case category:

Generative AI Operating Model & Roadmap

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A large Group and IM insurer partnered with EY to establish and rollout AI operating model, develop specialized ‘focus areas’ & use cases within each BU for Generative AI
activation, and assess the current state of their risk management policies and operating model.

Solution Provided
EY engaged in 30+ client interviews to understand current state risk management and operating models and outlined recommendations to stand up a new AI Leadership
Team & Interaction model, incl. change management considerations. EY developed a high-level roadmap for AI operationalization, including a timeline view of use case
execution over the next 24 months

Approach Taken (Phases and Key Activities) Value Delivered

 Leveraged EY internal knowledge to develop focus area selection frameworks and use  Gained alignment within both Group and IM BUs on 1) initial Generative AI
case prioritization frameworks focus areas and 2) use cases to be prioritized
 Held workshops and working session with BUs to identify focus areas (e.g., Small  Gained alignment across the enterprise on recommendations for the AI
Group and IM Life) Leadership and Governance structure
 Augmented existing use cases and identified new potential use cases; assessed each  Provided an inside-out view on the market landscape, including
use case against prioritization framework competitive view of what peers are doing / exploring in the Generative AI
 Provided recommendations on their risk management policies against industry-leading space
practices

Return back to meta data

 Use case category:

Generative AI and AI Lab

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client is looking for EY’s support with the following:  Defining a thorough Path to Production for AI project execution
 Establishing an AI Lab which is a central authority for AI-related initiatives  Setting up the AI Lab for use case execution and deliver
 Advising and guiding on key principles and leading practices for an AI program  Addressing inquiries regarding how AI technologies can benefit the business
 Establishing AI responsibility and ethics as a core tenant

Solution Provided Value Delivered

 Operational Efficiency: Promote organizational efficiency and effectiveness by  Defined purpose and values through the AI Lab Program Charter
defining operational materials for the AI Lab:  Structured Governance Model for effective decision-making and accountability
– Program Charter  Outlined the target state Operating Model for AI and Generative AI
– Governance Model  Established a clear Path to Production process for AI Lab use cases to reach
– Operating Model production
 Path to Production: Define Path to Production for an AI project that outlines roles  Implemented an Acceptable Use Policy to guide employees around AI and
and responsibilities for the use case intake and execution process Generative AI usage
 Developed a Risk Matrix Framework to address and mitigate risks associated with
 Legal and Ethical Considerations: Identify and address legal, compliance, risk, and AI and Generative AI
ethical concerns relevant to AI and Generative AI (e.g., IP, privacy, copyright,  Created a Third-Party Management Framework specifically for managing third-
transparency, explainability, bias, fairness) party relationships in AI and Generative AI
 Data Management: Facilitate data management discussions between the AI Lab and  Agreed-upon a Data Management plan to ensure compliance, privacy, and data
the Chief Data Officer (CDO) quality
 Accelerator Identification: Identify Microsoft accelerators to expedite the AI Lab's  Established an AI Lab Environment for exploration and experimentation
setup

Return back to meta data

 Use case category:

Generative AI Strategy & Governance for a Global Payments company

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The increased usage of AI Use Cases in recent times has caught the attention of regulators. We are going to see formal guidance from Regulators on a recommended mechanism to manage
the use of AI applications in financial institutions, and the associated risks. Currently there is no AI governance structure at the client.

Solution Provided
 Develop a governance framework to help organizations manage the risks associated with AI applications
 Address absence of a formal governance structure for AI applications
 Improve adequacy of controls around fairness and privacy aspects of AI applications
AI Governance Control Framework
AI Use Case
Deployment Framework Design Design
AI Use Case
AI Use Case AI Use Case AI Use Case
Performance
Initiation Development Testing Retirement
Risk Management Overall Program
AI Use Case Support
Change
Management

Approach Taken (Phases and Key Activities) Value Delivered

 Developed AI development and validation standards  Risk Reduction: Well-Structured AI Governance Framework
 Developed a sophisticated prioritization framework for implementation of AI use cases based on the risk – Streamline AI risk management associated with the development and
profile use of AI applications
– Developing end-to-end controls framework for all lines of defence, alongside their technology function
– Developing the AI Risk management Policy and Procedure for non-MRMP AI use cases, including  Risk Reduction: Effective Risk Tiering Framework
Generative AI ones – Assess risks associated with AI applications and facilitate
– Developing a comprehensive AI risk taxonomy that also includes idiosyncratic risks pertaining to implementation governance and controls
Generative AI use cases
– Developing a risk tiering approach effective in classifying AI use cases appropriately based on the risk  Time Savings: Reusable Templates and Artifacts
profile thereby informing the governance controls – Support operationalization of the developed governance framework for
– Establishing a target operating model and the associated RACI matrix for the AI use cases current AI model and any changes
– Assisting the client in integrating AI risk tolerance into their risk appetite

Return back to meta data

Section Generative
divider over two
AI:lines
GPS or
three lines

Page 245
 Use case category:

Funding Proposal Drafting and Management

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

What is the optimal balance of investment across the multiple determinants of health that will improve health EY Impact uses aggregates credible data to provide
outcomes and minimize disparities at the population level? How can a public agency maximize the funding available a detailed snapshot of health outcomes and
for community programs? The Federal grant application process can be resource-intensive and arduous, and public disparities within the state. The statistical model
agencies often lack resources to apply for grant funding that would make a critical impact in their community. identifies the regions and issues that need urgent
attention, driving resources to most impactful
Approach Taken (Phases and Key Activities) programs and providing a framework for outcomes-
based tracking. Insights are matched against open
Matchmaking & Application Human-in-the-Loop Fine- grants from Grants.gov to identify relevant funding
Tagging
Generation Tuning sources for current and new programs. With the
Funding Match grants with the state incorporation of Generative AI, EY Impact drafts a
Opportunity Topic 1 persuasive response to a Grant RFA, minimizing the
Topic 2 time spent and maximizing the competitive edge of
Topic 3 Display each section,
your application.
smart editor, etc.
Var 1 separately
Var 2 Value Delivered
Var 3 User edits each section
Extract Submission Format through pre-set prompts ► Improved efficiency: streamlines application
 Table of Contents & regeneration process and saves time
 Project Summary ► Consistency and accuracy: Minimizes errors and
Save – Background misunderstandings
Variables – Approach
Download grant proposal
► Customized suggestions: Generates tailored
– Etc.
draft & submit
proposal content
Match keywords from the grant Provide novel data-driven Auto-complete application format ► Learning from past applications: Informs future
opportunity with the connections between clients with the using function-specific models & strategies and best practices
corresponding health factors highest chance of receiving funding prompts, increasing output quality ► Capacity building: Enhances ability to secure
based on relevance scores and relevant grant programs and reducing error funding and address community needs

Return back to meta data

 Use case category:

EY Intelligence Engine Enterprise Asset Management Solution

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Legacy enterprise asset management systems face several challenges, such as the high lead time to get insights, limited scalability and integration, lack of real-time
visibility, and no proactive insights. EY’s Asset Management Solution seeks to solve these problems.

Approach Taken (Phases and Key Activities) Solution Provided

 GPT Use Cases: Fit for purpose storage and a central data model to
– Bot can analyze and answer data-specific questions across many systems/databases combine like data attributes and traverse data using
– Ability to contextualize instructions and take actions from Unstructured documents
Generative AI. Generative AI provides self service access,
aggregate, and consumption of data irrespective of
– Ability to link with other data sets to derive insights for intuitive suggestions
what disparate systems the data resides within.
– Text message based interactions with data enable real time and on the ground answers to questions
Solution enables users ability to automatic identify the same
assets between systems with different SKU and object
reference methodologies.

Value Delivered
 Common Data Model for Asset Management: Utilizing a
common data model across various
 Document intelligence and actions: Automatically parse
documents and service notes for assets into structured data
 Interactive map filters: Display real time asset data across
locations on a map for enhanced insights, filter assets
based on geolocation
 Integration with external systems using a data fabric
approach: Seamless integration across various systems
allows ordering parts with a simple click, and integrates
data and insights into a single pane of glass

Return back to meta data

Section Generative
divider over two
AI:lines
H&LSor
three lines

Page 248
 Use case category:

Generative AI Transformation for MedTech

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A prominent American MedTech company needs support in developing and implementing a robust Generative AI strategy to avoid falling behind in the industry. While they
do not aim to be the frontrunners, they recognize the necessity of staying current and cannot afford to lag behind their competitors. In addition to their ongoing data fabric
initiative and technology transformation program, the company aims to incorporate Responsible AI governance to enable innovation balanced with risk management.

Solution Provided
The AI Strategy project blends early deployment Generative AI use cases that can uplift revenue and operational opportunities in the near-term, while in parallel, developing
the organization’s medium to longer term AI strategy, roadmap and governance methods. Indicates which persona drives the
AI Strategy AI Portfolio Initiatives process step

The AI Strategy further details the high-level data architecture integrating the AI Hub and the Concept Review
Data Fabric.
DO DM PfM BO EA New Capability? DIL
New Acquisition ERP 1 ERP 2 Other
1 2 3 4 5
Data sources Sources Enterprise Submit New Validate Review and Digital
Data Demand Intake completeness validate/ refine
Conduct EA ye Innovation Labs
Warehouse Data Hub review
Form of the demand demand s Review
no
Next Gen Data & AI Engine
AI/ML Hub Promote to
Modern Data Platform (Harmonization Layer) Demand?

yes
Integrated Reporting Integrated Data End no Value-Analysis and
& Analytics & Document Catalog Portfolio Prioritization

Approach Taken (Phases and Key Activities) Value Delivered

 Define an AI strategy and generative AI operating model (define generative AI use cases)  Enabled our client to launch an enterprise-wide AI initiative, propelling their
 Establish a framework to evaluate initiatives based on various criteria (e.g. feasibility, AI maturity ahead of their peers and improving their competitive positioning
estimated ROI)  Build consensus and alignment for the client’s AI executive sponsor team (CIO,
 Define and stand-up ongoing governance strategy and pilot PPO value realization process CISO, Legal, Privacy and Risk officers) to establish a working AI governance

Return back to meta data

 Use case category:

CSR Generative AI
L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client wanted to create an end-to-end solution for safety section generation to improve efficiency by streamlining the CSR writing process.

Approach Taken (Phases and Key Activities) Solution Provided

3. Human-in-the-Loop The solution offers a user-friendly interface with human-in-
1. Extracting Data 2. Writing CSR Sections the-loop features, allowing detailed review, editing, and
Feedback
generation of report sections.
Map relevant data tables to
sections needing them  Permits section-by-section review, content re-generation,
and content editing with the ability to track user changes
Clinical Trial
Data in  Enables user to generate sections with or without
Document- Display each section, references for ease of review and table tracking
Based Tables smart editor, etc.
separately  Incorporates color coded evaluation metrics so the user can
easily understand how successful the generation was
User edits each section  Tracks progress throughout to remind the user of missing
Output and Evaluate
through pre-set prompts sections needed to review for ease of direction
& regeneration
 Clinical Lab Eval  Once all sections have been reviewed, users can download a
 Adverse Events
Brief Summary Download CSR & submit document file of the report (all sections)
Extraction and Save –
Transformation Results – Deaths
– Etc.
Write report, combining data in
Value Delivered
Auto-complete report using
Using data provided in different novel ways and cutting through the Creation of fully automated pipeline to generate a complete
purpose-specific models &
formats (e.g. HTML, PDF), extract noise. (Un)supervised auto-
structured tables evaluation for both training and
prompts, increasing output quality CSR, citing table references, inserting related intext
and reducing error via human input figures/tables, and generating applicable evaluation metrics.
production.

Return back to meta data

 Use case category:

Medical Legal Review (MLR) / Promotional Materials Lifecycle (PML)

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The MLR process and the broader PML are onerous and time-consuming, involving a significant amount of time by skilled reviewers, and the use of several external agencies. The business
team owning this process wanted to explore the potential of Generative AI to shorten and streamline this process and reduce their spend on external agencies.

Solution Provided
We built Generative AI and NLP solutions to handle a variety of scenarios within the business process:

Semantic search of approved Providing a function to translate An interface to identify changes Performing fact-checking by
documents that are stored in promotional materials from one in versions of documents, to comparing assertions made in the
Veeva PromoMats improving on language to another, accounting support the human reviewers and promotional material to the source
current keyword search function. for natural language patterns and prevent re-reviewing entire documents and validating whether
domain specific terminology. documents. those assertions are factual.

Approach Taken (POC Phases and Key Activities) Value Delivered

 Enhanced content discovery using AI-driven search,
making promotional materials easier and quicker to find
 Enabled English-to-Spanish translation of promotional
Define Test content, improving access to multilingual materials and
Setup tech environment AI model optimization speeding up readiness for new markets; turnaround
use cases and refine
reduction of 1-3 business days to 1-5 minutes
 Identify the specific task  Leverage client’s tech  Train the model on your  Test the model on a  Developed a feature to automatically detect and highlight
you want to accomplish (E.g. GPT4, Azure preprocessed data to separate dataset to content changes. This allowed content reviewers to focus on
using AI cognitive services etc.) improve its performance evaluate its performance.
changes, thus saving time and effort; turnaround time
 Select model options for your specific use case. This step is crucial to
reduction of ~40-80% per document
 Initialize provisioning via  Exposing APIs for ensure that the model is
Cloud portal external systems to working correctly and  Demonstrated the use of AI for fact-checking promotional
 Obtain base model, interact, generate and that it can handle new, content, decreasing reliance on external agencies by cross-
Identify integration API to process data unseen data verifying with authoritative data sources; turnaround
connecting systems reduction from 1-5 business days to 10-20 minutes

Return back to meta data

 Use case category:

Intercompany Agreements / Contracts analysis

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The Pricing Transfer team wants to extract and analyze the complex structure of intercompany agreements used to manage the relationship between the client's
subsidiaries and entities, enabling better identification of risk issues and more proactive prevention of risks such as tax disputes, resource inefficiencies, and suboptimal
capital allocation. The current process is primarily reactive due to the complexity of the agreements and capacity constraints.

Solution Provided
Developed Generative AI capability to facilitate semantic search functionality within contracts and conduct Q&A analysis on the content of multiple contracts. This advanced
system aims to enhance contract management by extracting key terms and highlighting anomalies. This has empowered the users with an intelligent tool that not only
enables efficient search within contracts but also provides a comprehensive understanding of contract content through advanced Q&A capabilities.

Build a searchable Identify and classify Extract key terms and Query the LLM to Highlight anomalies and
database of all contract types and anomalies from the unlock additional differences to analyze
contracts terms vector database search capabilities contract compliance

Approach Taken (Phases and Key Activities) Value Delivered

1. Created a searchable database encompassing different variations within contracts for ► Able to eliminate manual effort of legal contract reviews (estimated at
efficient retrieval and analysis. ~$600K) ensuring swift identification of anomalies and differences for
2. Identified and classified contract types and terms within the centralized database. comprehensive contract compliance analysis
3. Extracted key terms and anomalies from the vector database to enhance data ► Generative AI powered contract analysis reduces the time spent on
insights. document analysis and unlocks advanced search capabilities reducing
4. Utilized Language Model (LLM) for natural language user queries to unlock additional discrepancy between different intercompany arrangements (estimated
search capabilities for in-depth contract analysis. to reduce risk impacts by $millions on probability-adjusted basis) with a
scaled production solution
5. Highlighted anomalies and differences across contracts, enabling a thorough
examination of contract compliance for comprehensive assessment. ► Facilitated the automatic extraction of key terms and anomalies from
vector database, accelerating data insights and drill-down capabilities

Return back to meta data

 Use case category:

Commercial Market Landscape Reports

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The Commercial team conducts thorough research on potential new therapeutic areas or acquisitions of interest to the client. Their objective is to reduce the decision-
making timeframe by using Generative AI to automatically generate market landscape reports that are derived from credible internal and external sources of market
research data.

Solution Provided
Established a Generative AI capability to systematically search and retrieve market research data related to a specified topic from external market research databases and
sources. This involves a meticulous review of the returned search results to assess their relevance, with a focus on filtering out documents and information that lack
credibility or are not pertinent to the subject. The final step involves utilizing the identified and relevant market research data to generate a comprehensive market
landscape report on the given topic.
Search and retrieve Text extraction and Create Chunk Extract relevant data Generate structured
relevant external chunking from the embeddings from the points and generate market overview report
market research data relevant documents relevant data points initial report sections in PowerPoint format

Approach Taken (Phases and Key Activities) Value Delivered

1. Analyze source material from credible sources and potentially enrich it with additional ► Reduced the time to retrieve, select and analyze relevant source
material, including connecting to live data sources, to populate sections of the Market material from credible sources and internal databases for the Market
Landscape Report. Landscape Report from ~3-5 days to ~20-30 mins by leveraging
2. Generated a draft of a subset of the market report on the preferred topic, Generative AI text generation capabilities
encompassing natural language components. ► Eliminated the need of human input in generating Commercial Market
3. Created a .pptx file in a standard templated format for the Market Landscape Report. Landscape reports by automated content generation, re-generation,
4. Implemented a human-in-the-loop component in the final Market Landscape Report to review, and export of the CML report for the desired therapeutic area
ensure accuracy through review and regeneration of sections. ► The CML Report Generator product facilitates and ensures rapid analysis
of new business development areas and improved go to market speed

Return back to meta data

 Use case category:

Fortrea Data and AI Strategy

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A Clinical Research Organization (CRO) recently spun-off from a global clinical laboratory network. Post-spin, the CRO was struggling to maintain services and products as
foundational data platforms, cloud infrastructure, and talent were not included in the spin. The CRO client needed to assess the scope of capabilities and gaps in its data,
analytics and AI program in order to design a future state strategy to leverage AI and advanced analytics to support business objectives.

Solution Provided Value Delivered

 EY brought a cross-disciplinary team from our data, AI and analytics practice and from  Accelerated onboarding for the new data lead via current state
our strategy practice to assess the current state data and analytic program and make documentation and conceptual architecture
recommendations to inform the future state data strategy  Supported funding request and organizational model for the new AI team
 Our deliverable provided outside-in analysis of the market and inside-out assessment of via AI strategy documentation
capability against leading practice to provide recommendations on the path forward  Jump-started the data team with a backlog of business-critical data
 EY serves as a trusted advisor across the across the data, AI, and IT functions of the management, analytics, and AI/ML use cases
business and partnered to develop high level strategies for each function to support  Surfaced infrastructure gaps, risks, and dependencies across the
organizational planning and budgets enterprise to drive coordination in IT and data related initiatives

Approach Taken (Phases and Key Activities)

Strategic Alignment Current State Assessment External Market Research AI/ Data Strategy Design
• Align data and AI strategy to corporate • Canvassed the business for • Benchmark data and AI maturity across • Surface initiatives in-flight and highlight
strategy and define guiding principles data/analytics/AI priorities and gaps the market dependencies and gaps
• Review current documentation and • Inventoried data initiatives in-flight and • Survey customers and competitors to • Prioritize the backlog of data and AI use
define key customers, products, and highlighted risks identify priority use cases cases
services • Assess data management, governance • Inventory use cases for AI and analytics • Roadmap activities to action around use
and capabilities against leading practice to meet internal and external demands cases and scale capabilities

Return back to meta data

 Use case category:

Generative AI Powered Agronomy Assistant

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

► Can we use Generative AI to provide easy access to enterprise-grade knowledge  We were able to build a RAG-powered chatbot that outperformed ChatGPT
across a broad set of specialized agronomic areas to improve the knowledge base across all in-scope topic areas, providing value across sales, support & agronomy
of all sales, support & agronomists?  Solution went live to 100+ users that made a rich source of knowledge across
► Can we create a customized agronomy chatbot that outperforms out-of-the-box a breadth of topics easily accessible via a web application
ChatGPT?  POC proved value: EY is now working with the client to scale the application to
employees in North America by March 2024

Approach Taken: Iterative Feedback Loop Value Delivered

 Enterprise Data: Collected a rich knowledge base through enterprise datasets
Model Generate across a breadth of topics as well as multiple modalities such as text, audio &
Answers SMRs Score Scorecard
Model images
Answers  RAG: Utilized GPT4 to develop a Retrieval Augmented Generation (RAG)
solution that dynamically retrieves relevant internal & external client
documents to answer a user’s question
 SMR Feedback Loop: Leveraged SMRs by developing a question-and-answer
Scorecard
test bank, and then using SMR’s domain insights we iterated over the model
SMRs Write A B C D E
using prompt engineering or by adding more relevant data
Questions & E  Scorecard Evaluation: Developed a scoring methodology that can prove
Answers M outperformance compared to ChatGPT across not only a breadth of topics,
H but also for depth of a particular topic as well

Address Technologies
Data Gaps +
Source Prompting
QnA Bank Identify Gaps Azure
Azure Azure Azure
& Optimize Cognitive Streamlit
Search OpenAI Storage ML

Return back to meta data

 Use case category:

Generative AI - Conversational Business Intelligence

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

Business users rely on a small team of data scientists to get insights from the data. And a large portion of time for EY worked with the client to understand the current
data scientists are spent on providing insights and creating visualizations to ad hoc queries. The client seeks to state workflow, and delivered a proof of concept
use Generative AI to create a natural language solution that streamlines access to data insights and that allows users ask questions and generate
visualizations, thus improving efficiency and decision-making visualizations with natural language through a chat
interface
Approach Taken
Metadata Prompt Code Execution
Captured key data Prompts incorporate all relevant Engineered iteration loops to
Technology
definitions and common metadata and user request to enable tool to self-resolve errors  Microsoft Azure
business calculations return executable code and return successful code  Streamlit
 Plotly
User Input User Input / External Error
User Input
Data & Preference Output Interaction Handling
Text question
Selections

System Prompt LLM Output

Based on user
selections
Prompt Generation
LLM Interaction
GPT-4
Executable Code to
Query Dataset
Value Delivered
Code Failure  Data Democratization: POC demonstrated that
Sample Data Code Execution
2 lines of data from
Up to 5 follow-on
Runs locally to query non-technical users are now able to use tooling to
interactions to update
each in-focus table
code
full dataset interact with datasets without engaging data
Metadata: Final Output science teams to generate ad-hoc queries
Collibra - Data types & Query Shows query results +  Business Specific Language: Integrating business
Element Definitions summary to answer
Results
user query terms and calculation details in input metadata
KPI Detail: ensured that users can speak in normal business
Business Definitions &
Calculation Detail
Summarization Prompt
LLM Interaction Visualization language and receive accurate results from the
Gen. Query results +
user question
GPT-4 Generation model

Return back to meta data

 Use case category:

Generative AI - Standard Operating Procedure Navigator

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

The standard operating procedure documents define critical steps for day to day operations and it’s crucial for EY worked with the client to understand the pain
employees to execute their tasks according to SOPs. Currently, these documents are not easily accessible to points, the use case requirements for searching and
users, and locating the right section of the documents takes users a long time while they are required to work in a retrieving SOP documents and delivered a proof of
highly time sensitive environment. concept that allows users quickly get the accurate
SOP reference through a chat interface.
Approach Taken

Context Search Prompt

Technology
Developed a custom data Iterated through search Iteratively developed highly  Microsoft Azure
chunking strategy to ensure methodologies and returned specific prompt instructions  Streamlit
appropriate context and context to ensure all relevant to deliver a consistent result  Qdrant vector database
metadata could be extracted information is provided to the which met business user
from SOP documents stored model expectations
in PDF format Value Delivered
 Efficiency: The POC effectively addressed current
pain points in the SOP document searching process
Who is reducing search time from minutes to seconds
responsible  Enhanced Compliance: The SOP Navigator
for X? enhances the ability for employees to rapidly
access SOPs in the workplace, easing adhering to
compliance requirements and mitigating past
Knowledge Base Break information User Search for the LLM processes Generate challenges
of business into small single submits a chunks which are the user prompt response to  POC Capability Demonstrated: the POC provides a

procedure paragraph ‘chunks’ prompt most relevant to and context prompt clear path to production, which exemplifies high
documents and store as vectors the user prompt value realization in low complexity and low
implementation risk use cases

Return back to meta data

Section Generative
divider over two
AI:lines
P&U or
three lines

Page 258
 Use case category:

Generative AI Standards Bookshelf Search

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
A leading Power & Utilities company in the New England market relied on a legacy knowledge management system to store standards documentation. This method lacked
both a consistent method of indexing and an effective search capability. The client wanted a solution that could search the documents without relying on manual indexing.

Approach Taken (Phases and Key Activities) Solution Provided

The solution uses embeddings (numerical representations (vectors) of the meaning of  EY designed a solution that leveraged the advances in generative AI and
text) from thousands of documents. Users search for documents in plain text . an existing virtual agent to provide an Azure-hosted semantic search
A custom-built service integrates with Azure OpenAI to return the most similar results. experience.

Text scraped from  The solution solved the issue present in the legacy system used to store
documents and normalized documentation:
– A consistent method of indexing
– Effective search capability
Text embeddings created

Text embeddings stored

Text embeddings retrieved Value Delivered
 6,000 documents are now searched by meaning. This has led to:
– Improved employee experience
Search embeddings Better, more informed decision-making
Similar texts identified –
created
 Sematic search solved a key business problem while providing the value
of OpenAI, leaving behind a rich backlog of generative AI use cases
User enters search prompt

Return back to meta data

 Use case category:

AI Innovation Studio and AI Innovation Hub

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client is a leader of technology in the P&U space, and to keep pace with developments in AI in recent months wanted to stand up an AI Innovation Studio and Innovation Hub
within their IT organization. The Innovation studio was designed to enable use-case identification and development and to build out the organization’s AI capabilities further.

Solution Provided
EY used the Innovation Operating Model Framework to create the optimal operating model for the AI Hub in key dimensions across Leadership, Delivery and Technology.
Business, Strategy and Vision Drive AI-empowered innovation to enable business strategy
Establish governance processes that ensure ethical and responsible use of generative AI, including data
Governance, Risk and Compliance
privacy, security and transparency
Leadership
Financial Stewardship Achieve transparency of spend and value in AI solutions across the enterprise

AI Innovation Capabilities and Services Deliver AI services and capabilities for business growth at scale
Hub Operating Delivery Org Structure and Talent Optimize reporting structure for AI Hub and lead strategic workforce planning
Model Partnerships Collaborate to deliver new AI capabilities, enhance capacity and realize cost efficiencies
Framework
Architecture Streamline AI products, services and platform architecture
Technology
Stack Apps, data integrations and Security
Harness the power of apps and data while enabling a strong security foundation for ethical and responsible
use of generative AI, including data privacy and security
Establish resilient scalable, flexible and robust AI foundation including data and technology components
Infrastructure and Cloud
(foundational infrastructure and cloud

Approach Taken (Phases and Key Activities) Value Delivered

 Three high level pillars of the approach were Strategy, Execution and Governance  Faster time to market
 The Execution phase comprised of the Build, Operate, Transfer and Grow phases  Ability to test and refine the operating model before taking on operations
 EY applied an integrated AI lens to the Build, Operate, Transfer and Grow phases  Achieve scale, maturity and demonstrate value proposition in a short time

Return back to meta data

 Use case category:

Generative AI Governance, Platform Development, and Use Case Scaling

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client, an energy division of a large, multinational conglomerate, requested support in developing a comprehensive Generative AI strategy in order to successfully move
from a POC phase to a robust, comprehensive AI program will prove secure, efficient, and effective

Solution Provided
EY developed a 5-pronged approach to help the client succeed in their Generative AI program. This approach included Strategy, Platform Development, MVP Use Case
Development, Confidence in AI, and Workforce Upskilling

Strategy Equipping and Upskilling the Workforce

Help to rethink strategy and operations — from project intake, value creation Leverage AI to unlock the full potential of the client’s workforce, including
assessment, throughout the entire portfolio management process deploying new learning and upskilling programs.

Functional Transformation Confidence in AI

Develop MVP use cases. The 2 MVP use cases selected A) LLM to review terms Responsible AI frameworks and Cybersecurity controls to establish guardrails
and conditions of contracts for repairs and B) LLM for proposal generation and navigate an evolving dynamic legal and regulatory landscape

Building the foundation

Accelerate the adoption and integration of AI across the enterprise to support existing technologies, capabilities, products and services with a comprehensive
Generative AI platform

Value Delivered
▸ Safe, secure AI ▸ Workforce improvement operational ▸ Scalable platform architecture ▸ Quick win MVPs to prove value
▸ Comprehensive cybersecurity controls efficiencies ▸ Long-term Generative AI strategy

Return back to meta data

Section Generative
divider over two
AI:lines
TMT or
three lines

Page 262
 Use case category:

Enhancing Employee and Guest Experience

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
The client wanted to utilize EY to explore ways to generate value utilizing Generative AI to enhance their employee and overall guest experiences. The aim of this project
was to deliver three end-to-end proofs-of-concepts, architecture diagrams, and hand-off documentation with code and identified challenges of developing for production.

Approach Taken (Phases and Key Activities) Solution Provided

Using Generative AI and LLMs, the team developed applications to synthesize historical Generative AI solution using mock data and APIs within non-client systems
guest interactions into actionable insights, streamline knowledge retrieval, and automate to develop three end-to-end proofs-of-concept solutions. This production
feedback processes, all supported by generated mock data for a fictional theme park as a focused development centered around three use cases:
stand in for client data.  Summarization and insights to generate notes for guest Interactions
 Enable park staff to search internal business rules with natural language
Generated Artifacts
queries

Core theme park information Guests and events (emails, calls, interactions)  Process incoming email from guests, automatically tag, route and provide
suggested response template

Rides Hotels Concessions Category Topic

Value Delivered
 Enhanced staff workflows and guest experiences by integrating LLMs into
Number of
pre-existing workflows
Characters Ticket policies Mood/
style/ tone trips and  Demonstrated potential for greater process efficiencies
interactions
 Developed new methodologies for LLM output measurement and
benchmarking
LLM LLM generated LLM generated LLM generated  Begin continuous process to redefine real-time feedback and curate
generated interaction log call templates guest feedback
wiki pages emails unmatched, personalized guest journeys using Generative AI technology
in the future

Return back to meta data

 Use case category:

Marketing Compliance PoC

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement Solution Provided

A large technology company aims to streamline its marketing asset compliance verification by leveraging applied The PoC aggregates multiple inputs (partner
and Generative AI technologies. The current manual process of ensuring compliance with vendor guidelines for rulebook and creative ads) that are part of a
images, animated banners, and similar assets demands significant time and resources from the client. To enhance typical creative ad review. Utilizing GPT4V, the
efficiency and accuracy, the client is seeking to automate this procedure by integrating advanced technologies model first synthesizes large documents to
such as computer vision, OCR (Optical Character Recognition), and OpenAI vision models. consumable rules that the user can modify. With
the rules and images ready, the AI model utilizes a
Approach Taken (Phases and Key Activities) combination of GPT4V and traditional CV methods
to check against the provided images. The model
Input Processing Model Analysis Output and post-processing provides a simple compliance report for marketing
professional to correct ads (if applicable) and a
1 Rulebook 5 dashboard for further insights. The solution already
Compliance represents more than a 90%-time efficiency gain
Upload and
report and fewer manual creative ad reviews.
Digitization
4 Compliance
2 Custom rule check using 6 Compliance
ingestion Database Value Delivered
rules and
Image data  Improved efficiency: Represents immediate
3 Upload single 7 productivity gains and quicker analysis time
Compliance
or multiple  Consistency and accuracy: Minimizes errors rate
Analytics for
 Funding utilization: Allows to tap into full
images trend analysis funding for higher compliance
 Risk minimization: Reduces number of ewer
Provide the user optionality to Perform image analysis using Generate an output report and
dashboard to review results. audits and chance of sanctions on the
analyze images against a GPT4V and traditional CV methods
Additionally, there is a back-end SQLlite department
summarized set of rules, that have to perform all necessary checks dB to provide analytical datasets for  Futuristic approach: A solution that can handle
been synthesized using GPT4. against the rules trend analysis multiple partners, channels, and geographies

Return back to meta data

 Use case category:

Generative AI Product Knowledge Assistant

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Client sales teams face challenges with the immense amount of product knowledge they need to retain for various models and trims of vehicles available for each automotive manufacturer.
Similarly, customers have very detailed questions regarding vehicle features and specifications while making purchasing decisions.

Solution Provided
The AI assistant can answer questions about:
Product Knowledge
EY collaborated with the client to build a Product Feature Lookup Inventory Availability Copilot

personalized AI chat assistant for store sales team,

acting as a readily available expert for detailed Trim Specifications Equipment & Tech Packages
knowledge on vehicle models, features, specs, and
comparisons. The same chat functionality will be Model Comparison New & Retired Features
integrated into a customer-facing app to serve
customer inquiries while they are shopping in-store Manufacturer Warranty Performance & Safety
or online.

Approach Taken (Phases and Key Activities)

Apply feedback in
Align and ideate with Design user Ingest and index Prompt engineering Conduct
iterative
client stakeholders experience knowledge base and content chunking user testing
deployments

Value Delivered
1. Boosts Sales Productivity: Equips sales teams with up to date, easy-to-understand info, accelerating the sales process
2. Increases Efficiency: Generates model comparisons and deep-dives, freeing the sales team to focus on customer interaction
3. Improves Customer Experience: Enables the sales team to deliver tailored, knowledgeable service for every customer interaction
4. Delivers 24/7 Support: Acts as an always-available assistant, providing round-the-clock support

Return back to meta data

 Use case category:

Recursive Story Creation System

L4: Implemented with client
L3: Built for EY
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
Epics as written today are filled with inconsistencies, plot holes, and other such problems that can distract an audience from fully immersing into the story. The client hired EY to lead with the creation of a
personalized AI platform utilizing generative AI to assist in the creation of a highly contextualized body of knowledge inclusive of world-building, story, cannon, image and mechanic generation.

Solution Provided Frontend

UI (Chat and
Frontend Interaction Graph & Recommendation Enhancement Mapbox
► Users interact with the world-building app ► Leverages Neo4j for graph-related insights, Search)
predominantly via chat on the frontend analyzing relationships, offering suggestions, or
flagging inconsistencies
► Through chat, they can revise content and
images in real-time, powered by the LLM Agent ► A Recommendation Engine query layer interacts
with the graph database for additional story Query Layer
Agent-Assisted Interactions
consistency and coherence and to generate
► The LLM Agent mediates the chat, allowing users
to tweak generation prompts and
deeper insights and suggestions related to the Recommendation LLM Agent
story's components,
retrieve revised content or images Engine
► The Vector Database and Image Storage
► For content, the agent queries: Vector Database
periodically sync with Neo4j,
for text-based details, LLM Content Engine for
maintaining alignment across all storage.
generating new dialogues or lore, and for Backend Backend Backend
images, requests are sent to an endpoint using Map Functionality (Mapping)
(Query and Storage) (Generation)
StableDiffusionXL, and visual outputs are stored
► Allows visualization and update of locations
in Image Storage (e.g., S3)
using Mapbox, with a PostGIS database.
The agent will also be able to access map details Image Stable
from this backend storage Diffusion
XL
PostGIS
Approach Taken (Phases and Key Activities)
LLM
Phase 1: Phase 2: Phase 3: Neo4j Vector Content
Solution design and Image generation and Graph and geospatial Database Engine
architecture diagrams vector store setup database setup

Return back to meta data

 Use case category:

Generative AI Architecture for a Financial Co-Pilot for a Bleeding-

L4: Implemented with client
L3: Built for EY

Edge Manufacturer
L2: Built as a Prototype
L1: Proof of concept

Problem Statement
We have identified use cases below for potential improvements to enable light touch filing procedures for the SEC reporting team and legal team.

Solution Provided
 Identification and research of new disclosure items
 Refine current disclosures to align with current trends
 MD&A impact/trend disclosure analysis
 Identification of regulatory changes
 Benchmarking for risk factors analysis (user: Legal team)
 Higher disclosure risk identification per SEC comment letters

Approach Value Delivered

 Design phase for selected use cases  Cost and time savings: Reduction of direct business efforts
– Milestone: Solution design document systems and owners of APIs – Manual effort reduction & productivity gain
 Setup environment  Improved operational efficiency: Enhancement of research quality
– Milestone: Refined and deployed end- to-end technical architecture – Research content completeness enhancement
 Generative AI model optimization – Enhanced clarity for financial information & data interpretation
– Milestone: Optimized and grounded model with the internal and external data  Increased ROI: Reusability and transferability of AI Model
 Test and refine – Continuous enhancement of model precision with use
– Milestone: Test Scripts, Testing Results – Continuous consideration of comments and trends based on current SEC
 Deploy and monitor comment letters
– Milestone: Deployed model – Capability could be carried to other reporting, for ex., ESG requirements

Return back to meta data