AI for Business:

The opportunity
of a lifetime

Susmita Mishra
Brand Technical Specialist
Foundation Models
are bringing an
inflection point in AI... ...but how enterprises
adopt and execute will
define whether they
unlock value at scale
Foundation models establish a new paradigm for AI capabilities

Enhanced capabilities

Traditional AI models Foundation Models • Summarization

• Conversational Knowledge
Massive
external data • Content Creation
Training Tasks
Q&A • Code Co-Creation
AI1 Translation
3 Prompting
AI2 Pre
Sentiment
Key advantages
Analysis
Trained Massive
AI3
Foundation • Lower upfront costs through
AI4 Model less labeling
AI5 • Faster deployment through
Object fine tuning and inferencing
AI6 + Enterprise
Recognition
proprietary data
Fine
Tuned … • Equal or better accuracy for
Model Prompting multiple use cases
…
• Incremental revenue.
through better performance
• Individual siloed models • Massive multi-tasking model
• Require task specific training • Adaptable with little or no training
• Lots of human supervised training • Pre-trained unsupervised learning
up to 70% reduction
in certain NLP tasks
Building blocks of generative AI

ChatGPT But there is a Which will

inspired interest… bigger concept… change business

Large language model Foundation Generative AI

model

Transformer
Anything
Great that creates
at text new content

Unlabeled
data
Opportunity to unlock business advantage with foundation models trained across
the breadth of enterprise data

Physical Asset Business Business

Management Sensor/IoT Automation
Process
Data
Data

IT Ops & IT Cybersecurit Threat Management

IT Automation Data y Data

Molecular Foundation Geospatial

Sustainability
Data
Models Data

LLMs
Natural
Code
Language App Modernization
IT Automation

Customer Care
Digital Labor
Architecture model

Take advantage of both traditional and Next-Gen AI Models in watsonx.ai

Traditional AI Generative AI

Zero-shot Prompting Few-shot Prompting Data-driven Tuning

1000s — 1000000s labeled data points / task No labeled data needed 1-10 labeled data points / task 100s - 1000s labeled data points / task

Task Task Task Task Task Task Task Task

Model Model

Labeled Tuning data

Handcraft examples

Labeled Labeled
Training
Data
Training
Data
Pre-Trained Foundation model

Supports both conventional NLP tasks and new human + AI, iterative workflows

IBM Software | © 2023 IBM Corporation

Enterprise Considerations
Generative AI adoption considerations, inhibitors and fears
80% of business leaders see at least one of these ethical issues as a major concern
Explainability Ethics Bias Trust

48% 46% 46% 42%

Believe decisions made Concerned about the Believe that generative Believe generative AI
by generative AI are not safety and ethical AI will propagate cannot be trusted.
sufficiently explainable. aspects of generative established biases.
AI. Agree Neutral Disagree

Enterprises need more than an AI solution - they need a comprehensive and sound strategy for generative AI.

Source: IBM IBV “generative AI: The state of the market”, June 2023
IBM POV: Four core principles to tailor generative AI for enterprise

Open Targeted Trusted Empowering

→ Based on the best AI → Designed for targeted → Offering security and → On a platform to bring
and cloud technologies business use cases, that data protection. your own data and AI
available. unlock new value. models that you tune,
→ Built with governance, train, deploy, and govern.
→ Giving access to the → Including curated transparency, and ethics
innovation of the open models that can be tuned that support increasing → Running anywhere,
community and multiple to proprietary data and regulatory compliance designed for scale and
models. company guidelines. demands. widespread adoption to truly
create enterprise value.
Introducing…

Watsonx
The platform for AI value creators
The platform watsonx.ai watsonx.data watsonx.governance
for AI and data Train, validate, tune and Scale AI workloads, for all Accelerate responsible,
deploy AI models your data, anywhere transparent and explainable AI
workflows

watsonx A next generation enterprise

studio for AI builders to train,
Fit-for-purpose data store, built on
an open lakehouse architecture,
End-to-end toolkit for AI
governance across the entire model
validate, tune, and deploy both supported by querying, governance lifecycle to accelerate responsible,
Scale and traditional machine learning and and open data formats to access transparent, and explainable AI
new generative AI capabilities and share data. workflows
accelerate the powered by foundation models.
It enables you to build AI
impact of AI with applications in a fraction of the
trusted data. time with a fraction of the data.
What IBM offers

IBM approach for AI: Unleash the intelligence in your business

Consulting Center of Excellence for Generative AI and Client Engineering for watsonx System Integrators,
for AI Software, and
SaaS partners

AI products Digital Labor IT Automation Security Sustainability Application

Modernization

AI and data watsonx

platform
watsonx.ai
watsonx.data
watsonx.governance

Hybrid cloud Red Hat

platform
OpenShift AI
Ansible Lightspeed

zSystems
Infrastructure AWS/Azure/Other
Distributed Infrastructure
for AI
IBM’s AI is based on foundation models that are multi-model on multi-cloud

Build your
own models

13

Use IBM + open Build and

Try Tune Deploy*
+ other models Evaluate

*on any cloud

BYO
Use IBM models Model
What IBM offers

IBM’s AI is empowering: it is
watsonx
You have multiple entry points
for value creators, not just users

Bring your own data

Bring your own AI

Use the platform to

>75%
tune, train, deploy,
and govern your AI

Models will be key IP for enterprises

Do not be limited to just prompting
are looking to fine-tune open-sourced someone else’s AI model with
models or build their own models no control over the model or the data
to fulfill their needs for governance,
latency, and regulatory requirements Own your models. Control your AI destiny.
What IBM offers

watsonx.ai
Train, validate, tune and deploy AI models

A next generation enterprise studio for AI

builders to train, validate, tune and deploy
generative AI, foundation models, and machine
learning capabilities

• Foundation Model Library with IBM and open-

source models

• Prompt Lab to experiment with foundation

models and build prompts for various use cases
and tasks

• Tuning Studio to tune your foundation models

with labeled data

• Data Science and MLOps to build machine

learning models automatically with model training,
development and visual modeling
Foundation Model Library Model variety to cover
enterprise use cases and
compliance requirements

IBM models Opensource models

IBM’s suite of foundation models are designed to ensure model trust Experiment with opensource
and efficiency in business applications. Our suite of models feature: models
IBM and Hugging Face
partnership demonstrates
our shared commitment to
delivering to clients an open
ecosystem approach that allows
Transparent Pre-Training on IBM’s Compute-Optimal Model Training and Efficient Domain and Task them to define the best models
trusted Data Lake Architectures Specialization for their business needs.

– Largest known repository of – Granite Models Coming Soon:

enterprise-relevant training data Decoder only transformers
– granite.3b.finance
– We've built and carefully vetted – Sandstone BYOB models (optional add on)
– granite.3b.cybersecurity
one of the largest business Encoder-decoder transformers
relevant data sets to train our – And more! (legal, climate, etc.) Partner with IBM Research to
– Obsidian (in progress)
models based on principles of trust Sparse universal transformers pre-train your own
and transparency
foundation models.
– Full, auditable data lineage
available for any IBM Model
watsonx.ai Foundation Model Library
Model variety to cover enterprise use cases and compliance requirements

IBM Foundation Models Open-Source Large Language Models

Slate (encoder-only) NLP models Encoder/decoder & decoder-only Large Language Models available in Prompt lab
(Fine tuning NOT required for most tasks)

Slate flan-ul2 gpt-neox mt0-xxl flan-t5-xxl mpt-instruct2

multilingual 20 billion 20 billion 13 billion 11 billion 7 billion
distilled encoder/decoder decoder only encoder/decoder encoder/decoder decoder only
153 million

Q&A Q&A Q&A Q&A Q&A

Fine Tuning Required to support:
Generate Generate Generate Generate Generate

Extract Extract Extract Summarize

Classify Summarize Summarize Classify

Classify Classify

Note: Slate models are fine-tuned Open-source models are sourced from Hugging Face
via notebooks + API

Q&A Model responds to a question in natural language Extract Model extracts entities, facts, and info. from text Classify Model classifies text (e.g. sentiment, group, etc..)

Generate Model generates content in natural language Summarize Model creates summaries of natural language
The most common Retrieval-Augmented Summarization Content Generation
Generation
generative AI tasks
implemented today Based on a documents or Transform text with domain-
specific content into
Generate text content for a
specific purpose.
dynamic content, create a
chatbot or question-answering personalized overviews that
feature. capture key points.

Building a Q&A resource from a Conversation summaries, Marketing campaigns, job

broad knowledge base, providing insurance coverage, meeting descriptions, blog posts and
customer service assistance transcripts, contract information articles, email drafting support

Named Entity Insight Extraction Classification

Recognition
Identify and extract essential Analyze existing unstructured Read and classify written input
information from unstructured text content to surface insights with as few as zero examples.
text. in specialized domain areas.
Sorting of customer complaints,
Audit acceleration, SEC 10K Medical diagnosis support, threat and vulnerability
fact extraction user research findings classification, sentiment
analysis, customer segmentation
watsonx.governance

Enable
responsible,
Govern across the AI Manage risk and protect Adhere to regulatory
transparent and
lifecycle by automating reputation by automating compliance by
explainable AI
workflows and consolidating tools, workflows to better translating growing
applications and detect fairness, bias and regulations into
An end-to-end platforms. drift. enforceable policies.
toolkit
encompassing
both data and AI
governance

Comprehensive Open Automatic metadata

Govern the end-to-end AI Support governance of and data
lifecycle with metadata models built and deployed in transformation/lineag
capture at each stage 3rd party tools. e capture though
Python notebooks.
IBM watsonx.governance integrates and augments your existing ML
development and deployment tools and processes

Model inventory Model risk governance

(capture model facts throughout
Sync model status (workflows, dashboards, incident
the lifecycle) and metadata management)

Capture
training
Capture
meta-data
deployme
nt meta-
data

Build Deploy approved model Deploy

(IBM, AWS, MS, Other) (IBM, AWS, MS, Other)

Capture model
performance
meta-data

Pre-deployment Post-
model evaluation Evaluation and monitoring deployment
(accuracy, drift, bias, model
explainability)
monitoring
Banking/Financial
Markets
The impact of generative AI

Banking/Financial Service Desk Amplified Financial Report

Services Use Cases Expertise Customer Care Summarization
with watsonx.ai
Pilot Program

A financial services company When customer care reps chat In order to make well-founded
has a large volume of complex with customers, they need time lending and M&A
service requests that to research and craft responses. decisions, bankers must research
require expert agents to and understand vast amounts
research multiple long-form Result: An AI chatbot sits quietly of financial reports and publicly
sources to address. This leads in the background of customer available information.
to an extensive backlog and conversations to suggest auto-
high ticket re-open rate. generated answers for reps from Result: AI generated summaries
an internal knowledge base. of SEC filings, i.e., 10-K
Result: Agents at any level of This enables more accurate reports, Public Information Books
expertise can resolve issues responses up to 65% faster, (PIBs) for bankers to leverage
using a while maintaining personal yields faster, better informed, and
natural language interface that touch. more consistent decisions.
pulls answers
from API, product or
technical documentation.
The impact of generative AI

Banking/Financial Assess Customer Automate Simplify Regulation

Services Use Cases Sentiment Workflow Compliance
with watsonx.ai
Pilot Program

Self-evaluation of Financial Services company Managing risk and compliance in

customer service performance has complex, complex regulatory environments
can be inaccurate with self- manual processes needed to involves a lot of manual work,
selection bias, and difficult to service private equity funds. which has potential for inaccuracy.
navigate interactions from
multiple systems. Result: Complex legal terms in Result: AI analyzes policies to
partnership agreements are identify gaps and inefficiencies in
Result: Bank using AI to extracted using watsonx.ai into order to minimize obligations
read customer conversations, a structured format with public regulations, policies
classify them into products, that supports automatic and controls and identify where
and gauge NPS/customer processing to save time and gaps exist.
satisfaction for actionable reduce manual labor.
insights.
The impact of generative AI

Banking/Financial Faster RFI/RFP

Services Use Cases Responses
with watsonx.ai
Pilot Program

Wealth Management team

puts in a lot of time-consuming
effort to sift through hundreds
of pages of RFIs/RFPs
to extract crucial details
for summarization documents.

Result: Document summaries

are now created with the click of
a button, resulting in a
substantial time reduction for
senior analysts reviewing,
mapping and summarizing
RFI/RFP documents.
Three ways to get started with watsonx.ai today
IBM’s investment in partnering with you

FREE TRIAL CLIENT BRIEFING PILOT PROGRAM

Experience watsonx.ai and test Discussion and custom demonstration watsonx.ai pilot developed with
out core capabilities yourself of IBM’s generative AI watsonx point- IBM AI engineers. Prove watsonx.ai
with a free trial of-view and capabilities. Understand value for the selected use case(s)
how watsonx.ai can be leveraged in with a plan for adoption.
your AI strategy.

Try our free trial 2-4 hours 1-4 weeks

Onsite or virtual