Microsoft Copilot

Studio
HANDS-ON WORKSHOP

© Copyright Microsoft Corporation. All rights reserved.

 Introduction to Microsoft Copilot Studio

Lab 1: Create your first Copilot in Microsoft Copilot Studio

Lab 2: Microsoft Copilot Studio: Authoring 101

Lab 3: Build and Invoke Power Automate cloud flows

Agenda Lab 4: Make HTTP requests to connect to an API

Lab 5: Knowledge sources, AI knowledge, and custom

instructions

Lab 6: Use generative AI orchestration to interact with your

connectors

Lab 7: Invoke AI Builder prompts

Q&A
Introduction to Copilot Studio
 What is a copilot?

An experience using generative AI to assist humans with

complex cognitive tasks.

Chat using natural You determine,

Its value increases
language (and guide, and approve
with complexity
code) the output
Microsoft Power Platform
The low-code platform that spans Office 365, Azure, Dynamics 365, and standalone
applications

Power Apps Power Automate Power BI Copilot Studio Power Pages

Application development Process automation Business analytics Intelligent copilots External websites

Data AI Builder Microsoft Power Fx Managed

connectors Dataverse Environments
 Microsoft Copilot Studio
Your copilot, your way

Copilot Studio is an end-to-

end conversational AI product
for
building your own
copilots or extending
Microsoft Copilot with
generative AI, large language
models and your data.
Different building journeys for different needs

Copilot Studio

Included in Copilot for M365 Standalone Azure AI Studio

Customize Microsoft Copilot Build your own custom copilots

Connect with your data sources Build customer or internal facing copilots with your data and workflows,
outside of Copilots reach on channels that work for you

Rules-based dialog management

Call your redesigned business Configure LLMs with prompt
and conversational orchestration
processes and workflows engineering
with a prebuilt LLM

Configure Copilot responses to Managed SaaS, no infrastructure, Manage own infrastructure,

specific questions built in security and governance custom dev and security

Build Conversational
Building Copilot Build Models
Orchestration
Extensions

Extend Copilot Studio

with Azure AI Studio
The end-to-end Copilot Studio Dialog Management

copilot
conversation
Authored Topics
STAGE 1
Unified Copilot-assisted
Authoring Canvas Authoring

1. Authored Topics
Organizations control their critical
topics by designing their own
specific processes and workflows. Generated Responses
STAGE 2
Generative Answers /
GPT Models
2. Generated Responses Actions
Generative AI answers queries at
scale that may be duplicative or
less complex.
Escalation
3. Escalation STAGE 3
If the copilot can’t handle the Omni-channel Live Agent Hand-off
conversation, it will escalate the
conversation to a human assistant.
 Copilots and Conversational AI
Build and extend across the Microsoft ecosystem

Custom copilots Microsoft 365 Biz Apps & Power Platform Other Microsoft Copilots

Build Custom copilots and bots Extend Microsoft 365 Copilot Extend Copilot experiences in Extend other Microsoft Copilot
for enterprises and third parties Dynamics 365 and Power Products Experiences

Copilot for Copilot for Service Power Platform

Sales

Microsoft Copilot Studio

Build custom Copilots | Extend and customize 1st party copilots

Bot
Bot Service Azure Azure Cognitive Power Platform AI Builder
Framework /
Channels AI Studio Services Connectors
SDK
Different building journeys for different needs

Copilot Studio

Included in Copilot for M365 Standalone Azure AI Studio

Customize Microsoft Copilot Build your own custom copilots

Connect with your data sources Build customer or internal facing copilots with your data and workflows,
outside of Copilots reach on channels that work for you

Rules-based dialog management

Call your redesigned business Configure LLMs with prompt
and conversational orchestration
processes and workflows engineering
with a prebuilt LLM

Configure Copilot responses to Managed SaaS, no infrastructure, Manage own infrastructure,

specific questions built in security and governance custom dev and security

Build Conversational
Building Copilot Build Models
Orchestration
Extensions
Our Focus today
 Build &
Publish

Copilot Studio

Analyze &
Improve
 1
Start Create a copilot
Stand up Copilot Studio’s SaaS in minutes.

Design the conversation

2
Build Connect your knowledge base for generative
answers, create rule-based topics, integrate

How to build a plugins and data connectors, and extend with

Azure AI Studio.

custom copilot 3
Publish to different
Publish channels
Publish your copilot to Teams, websites, apps,
social media channels and more.

4 Control and improve

Manag Monitor, manage and improve the copilot with
e enterprise-grade security and controls.
Build &
Publish
Create a copilot
Build a copilot and go live
within minutes, all from
one easy-to-use, E2E SaaS
product
Build &
Publish

Add knowledge
Connect to your
enterprise backends
and APIs with 1000s
of pre-built data
connectors
Build &
Publish

Customize topics,
and actions
Mix generative responses
with step by step topics
for complete control.
Create actions, plugins
and connect to your
backend systems with
1000s of pre-built
connectors
Build &
Publish

Publish and
go live instantly
Publish and deploy to
the channel of choice
with a single click,
and your copilot is
live.
Analyze &
Improve
Monitor and
get insights
Understand what’s working
well and what needs to
improve
with conversational analytics
Analyze &
Improve

Fine tune to add

sophistication
Layer in more topics with
custom responses and
conditions based on insights
generated to improve your
copilot’s performance.
Analyze &
Improve

Add Conversational
Services
Integrate with Azure AI
Studio, Azure Cognitive
Services, Bot Framework
and a variety of other
Microsoft conversational
services
 Create, manage, publish and extend
Copilots
Live in minutes - all from one tool
and E2E SaaS service

Integrate with AI Chat over knowledge with

Services Gen AI
Integrate with Azure AI Get enterprise specific answers
Studio, over your files, websites, internal
Azure Cog Services, Bot
Framework and various other Build & shares, Dataverse, third party
systems and more
Microsoft conversational
services Publish
Create specific topics
Supplement generative
Monitor and Improve responses with specific,
with rich out-of-the-box curated topics where you
insights and analytics want tight control. Build
them easily with the
Copilot Studio powerful graphical studio

Publish to multiple Build actions &

channels, and go live Plugins
instantly on the SaaS service
or choose to extend Copilot
Analyze & Create actions, plugins,
use 1000s of pre-built
connectors or Power
for Microsoft 365 with your
custom capabilities Improve Automate to call your
backends and APIs
Example use cases to build copilots today:
B2C B2B B2E

Customer Support Supplier Interaction Human Resources

Build your own

Handle routine customer Facilitate communication Interface with your ERP/HR
inquiries, providing instant between businesses and their systems to simplify processes.

copilot
responses and freeing up suppliers, managing orders and Ask your copilot
human agents for more track shipments.
complex
Ask issues.
your copilot Ask your copilot Start the onboarding process and

required tasks for new colleague

How do I return a product? What’s the status of our latest

How do I reset my account component order?

information?
Expense
Management
Interface with your ERP systems
Products/Services Lead qualification

Extensibility
to streamline your finance or
Assist customers with product Qualify leads by asking relevant resource planning processes.
and service discovery based off questions and then route them Ask Copilot
your website. to the appropriate sales
Ask your copilot representative.
Ask your copilot Send me a list of pending expenses

from this week.

What is the latest laptop? Can I fit Can you provide me with
An employee can have a Copilot for M365
it in my 15” backpack? information on your bulk pricing for license and still want a custom copilot.
office supplies? Audiences typically without Copilot for M365
like frontline workers.
Generative AI is changing bot building Traditional (e.g. Azure)

• Create Azure tenant and infra

• Annotate thousands of utterances

and conversations to build a
custom NL model

• Create topics in code using Visual

Studio and BF SDK

• Host Azure Bot Service and

Manually authored topics instantiate the appropriate
Anticipate and build a manual topic for channel
each possible user question
• Deploy your bot to the service
CRITICALITY

• Build your own custom analytics

BUSINESS

pipeline

Missed opportunity • Every time your knowledge

content changes (e.g. different
hours) go and update the bot in
code

Time spent Conversations

hand-building missed or
topics escalated
TOPICS
BUILT
Generative AI is changing bot building Copilot Studio with
Generative AI
• Full E2E SaaS – no infra to host or
manage

• No explicit NL training or custom

model required, comes with
powerful large language model
Manually author business • Point to a knowledge sources and
critical topics quickly with have a useful bot in minutes with
Copilot and Copilot Studio’s Generative AI
intuitive builder
• Extend the bot with business-
critical topics using easy to use
graphical interface and 1000s of
pre-built connectors, or call
CRITICALITY

custom APIs
BUSINESS

• Developers, low-code users can

Expose your knowledge sources collaborate inside the same
and APIs and generative AI canvas with graphical multi-
Custom handles the rest authoring and code side-by-side
build
topics • Publish to the channel of your
+ help choice in minutes with a few clicks
from Topics handled by
Copilot Generative Answers and
Actions
TOPICS
BUILT
 More than 30,000 customers across every industry
have used Copilot Studio
to help improve performance and efficiency while reducing costs and risks

Conversational banking HR/IT copilot for employees Helping customers find and Creating customer focused Customer copilot to help Copilot that helps with
copilot on both that reduced support costs book the perfect cruise copilots as a Microsoft find the right products and customer service for
telephony and digital and workload Partner support guests
channels
Copilot Studio
Copilot Studio
message capacity add-on
$200 per tenant/month
Tenant-wide capacity license sold in capacity packs.
1 capacity pack = 25,000 messages/month

Offer Details
• Gen AI: AI-enabled conversations
• Copilot in Microsoft Copilot Studio: Create and iterate on chatbot
topics using expanded natural language capabilities
• Power Automate for Microsoft Copilot Studio: Cloud flows – Create
automated, instant and scheduled flows within context of Microsoft
Copilot Studio bots
• Connectors: Standard, Premium and Custom connectors
• Dataverse for Microsoft Copilot Studio: Dataverse Default Capacity
o
5 GB Database Capacity
o
20 GB File Capacity
o
2 GB Log Capacity

All pricing subject to change, please consult price list for actual pricing.
Copilot Studio
Implementation Guide
The Copilot Studio implementation guide
provides a framework to do a 360-degree
review of a copilot or bot project.
Through probing questions, it highlights
potential risks and gaps, aims at aligning
the project with the product roadmap, and
shares guidance, best practices and
reference architecture examples.

aka.ms/CopilotStudioImplementationG
uide
Get Started
Go to
aka.ms/CopilotStudioWorkshop
THEN:
1. Download and unzip
the lab files locally.

2. Using the provided

SCAN ME credentials, log in aka.ms/
CopilotStudioStart
IMPORTANT LAB INSTRUCTIONS

1. Follow the labs most lab content is dependent on

previous labs.

2. If you want to deviate from the labs and test

your own scenarios, this is fine, but please do
complete the labs first, and create your own
separate custom copilot for your own scenarios
and exploration after you’ve completed the labs.
Productivity tip #1: Unpin the Copilots pane
Productivity tip #2: Refresh
the
conversation between tests
Lab 01: Create your first copilot
in Microsoft Copilot Studio
Demo Lab 1
Copilot for Copilot Studio

AI assistance in building topics,

designing and modifying the
copilot all through natural
language
Assisted authoring for:
• Topic creation
• Topic iteration
• Response generation
• Adaptive Card generation
• Topic improvement suggestions
• Suggested trigger phrases, topic names,
topic descriptions
• Transcript generation
• Copilot creation
• Topic suggestions
Azure bots Prompt-authored
Copilots took
take copilots will take hours
days to
months to or even minutes to
author
author author.

Generally Available
Lab
time!
Lab 01 [40]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Lab 02: Authoring 101
Demo Lab 2
Copilot Studio authoring canvas

Rich Power Fx-based Code Event-driven & Productivity

Responses variables system side-by-side contextual
triggering • Cut, Copy, Paste
New features including: • Power Fx integrated • Move between a • Trigger an event that
formula editor for visual editor and code can interrupt Dialogs • Multi-node selection
variable management • Share and re-use if it detects a certain
value has been • Enhanced topic
• List / Record type topic logic between navigation
received
support developers and bots
• Add more nuanced
• Condition nodes with
Adaptive Images responses to a
Cards Power Fx formula
message when the
support
bot runs into an issue.
• Iterate over a List
• Bots can stay within
with new Loop nodes
context when a user
(coming soon)
asks for additional
Videos Quick Replies help or more details
Lab time!
Lab 02 [60]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Disambiguation with topic design
Natural Language Understanding “Trigger Phrases” Topics Reusable, Bite-Size, Topics
Trigger phrases Triggered from trigger phrases or redirects Explicitly called by topics

Topic: Card
Topic: Debit Card

• Issue with my card Question: Debit/Credit

• Block card …

• I need to block my credit Question: Block/Unblock

card
• Need help blocking debit Condition
card Topic: Credit Card
• Unblock card
• Need to unblock credit card
…
• How to unblock my debit Debit Credit
card
End of Conversation

Topic: End of
Conversation

…
Topic: Location

• Find my nearest bank

Question: Location
• Check branch location
• Find a bank
• Location Topic: Location Check
Find me your nearest
location
• Banks near me Message
…

End of Conversation
Natural language understanding
Custom Azure CLU
Standard NLU model Generative AI orchestration
model

 Default, out-of-the-box, model that  Supports additional languages, with  Uses a GPT large language model.
comes pre-trained, with many native models.  Can handle multiple intents and chain
predefined entity types.  Allows to further customize the intent topics and/or plugins.
 Configuration is done by adding trigger triggering model for better intent  Automatically generate questions for
phrases and custom entities (either recognition or to address specific missing inputs and answers complex
closed lists with values and synonyms, industry requirements. entities and questions from the
or regular expressions).  Allows for complex entity extraction conversation context.
(e.g., of the same type).  Configuration is done by simply
 Entity extraction can also leverage describing topics, plugin actions, and
Copilot Studio standard NLU. inputs/outputs.

 Single intent recognition per query.  Single intent recognition per query.  As it’s a generative AI feature, the
 Cannot be extended.  Configuration is done in Azure and licensing burn rate of messages is
involves additional costs. higher than in regular topic triggering.
 Slot-filling multiple entities of the same
 Has its own service limits that need to  Preview capability.
type in the same query requires
disambiguation for each (e.g., from be evaluated.
and to cities)  Azure CLU intents and Copilot Studio
topics must be carefully kept in sync.
Handling unrecognized intents
Answering for unplanned user queries

1 3
The fallback topic gets triggered when If knowledge sources are enabled on the
Copilot Studio doesn't understand a user copilot, the conversational boosting topic
utterance and doesn’t have sufficient also triggers on the unknown intent event
confidence to trigger any of the existing and triggers before the Fallback one.
topics.
2 4
There are many ways to handle While it’s important to leverage the
unrecognized intents: using the AI’s own Conversational Boosting and Fallback
general knowledge, or knowledge sources to capabilities, it’s also important to make sure
ground the answer in your own data sources that the core scenarios and topics of your
and/or using the Fallback topic to integrate copilots are property handled through
with any other systems. custom topics and their outcomes defined
For example, question answering in Azure AI (resolved, etc.).
for Language allows you to offload large
volumes of question-and-answer pairs. It also
has a chitchat model to handle random
questions to the copilot.
Lab 03: Build and call Power Automate
cloud flows from your copilot
Demo Lab 3
Integration patterns considerations
Power Automate HTTP requests Bot Framework
Cloud flows & Connectors Skills

 No-code / low-code.  No-code / low-code.  Synchronous execution.

 Clear separation of integration and  Faster runtime execution.  Support for private endpoints.
conversational logic.  Can be monitored as part of Copilot  Existing Bot Framework investments
 Can be monitored separately. Studio App Insights integration. can be reused.
 Existing cloud flows can be updated to  Can leverage variables, including
integrate with Copilot Studio. environment variables and secrets.
 Variables, conditions, parameters, etc.
can use Power Fx formulas.
 Parsing and error handling support.

 Invocation and execution of the cloud  Mixing conversational logic and  Pro-code (e.g., C#)
flow can add latency. integration logic, but integration topics  Runs in the Azure AI Bot Service.
 If scale/performance are a concern, the can be isolated, as they can be
configured for inputs and outputs.  Additional costs need to be covered by
higher-tier “Power Automate Process” an Azure subscription.
plan can be evaluated.  Need to return results within 100s.
 ALM differs from Power Platform.
 “Power Automate Process” required for
service principal ownership.
 Need to return results within 100s.
Power Platform Connectors
Multi-Channel
• Power Platform, including
Power Automate and Copilot Website
Studio, offers great
integration capabilities, with
Mobile
more 1,200 native
connectors or ways to
create your own custom Teams
connectors to your APIs.
• For a good end-user Facebook
experience, cloud flows,
HTTP requests and Power
connectors triggered from Automate Telephony
Copilot Studio must execute
quickly so that the user Email
doesn’t have to wait too
long for the copilot to
answer. Slack

Copilot Direct line Speech

Studio

Many more

1200+ prebuilt data

connectors
Lab time!
Lab 03 [60]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Lab 04: Make HTTP requests
to connect to an API
Demo Lab 4
Lab time!
Lab 04 [20]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Lab 05: Knowledge sources, AI
knowledge, and custom
instructions
Demo Lab 5
Knowledge sources

Enable multi-turn chat over your own

internal and external knowledge
sources and sites with generative
answers
Copilots can answer thousands of
questions out of the box in seconds

You can even pull data from an API or

other backend system and enable
generative chat over it

Try now at
aka.ms/CopilotStudioDemo
Generally Available
 •Up to 4 URLs, but the list can be dynamic.
Knowledge • Websites must be
Public
Websit
•Max 2 subpages depth (/en/help/), no direct
pages.
•One configuration ID, but it can be set using a
indexed by Bing. e
sources & Public data • Bing being a global
service, no regional
Bing
Custom
formula.
•Azure costs covered by Microsoft

Generative AI boundary is •Up to 400 URLs, custom ranking options.

possible. •Max 2 subpages depth (/en/help/), supports
1 Query rewriting Search
direct pages.
Optimizing the user • Requires the user to be authenticated with Entra ID to make delegated calls to
question for search
the Graph API.
Last 10 turns are SharePoint
• Matching files (max 3 MB) are retrieved to get detailed snippets to summarize.
leveraged for Internal
• Up to 4 SharePoint or OneDrive for Business sites (but the list can be dynamic).
contextualization only
• Security trimming: returned results only include content the user has read
access to.
• Users eligible to Microsoft 365 Semantic Index benefit from better results.
2 Content •
Azure OpenAI Returns results from the linked Azure AI Search index (keyword, semantic, and
retrieval
Querying each source vector search).
on your data • Not a passthrough to Azure OpenAI – generative answers summarization and
Top 3 results per preview
validation apply.
source
• Connection isn’t delegated: no security trimming, no authentication requirement
• for the user.
Files (max 3 MB) are stored in Dataverse file storage and indexed with Dataverse
3 Summarization
Files
• Can index content from public URLs/web addresses (useful if regional boundary
Search.
Answer summarization • is required).
Not possible to offer a URL to the source documents in the citations.
from retrieved content • Documents are uploaded at the copilot level. Gen Answers node will search
across all uploaded documents or none, when disabled at a node level.
Citation generation
• Requires a step to query the data source – using cloud flows, connectors, or HTTP
Answer personalization
requests.
with custom Custom data • Results are passed as inputs to the generative answers to summarize an answer
instructions
for the query.
• Input data needs to be in table format, with 3 properties: Content (typically
Validation at each snippets of relevant content) , ContentLocation (optional, typically a URL) and
Moderation
step of harmful, malicious, Dataverse • Title
Dataverse tables (max 15).
(optional).
uncompliant, or copyrighted preview • Natural language queries are transformed in analytical queries (e.g., SQL) over
answers structured data
Grounding validation and
Graph • Connect to additional enterprise knowledge sources that are indexed in Microsoft
hallucination removal
Connectors Graph.
preview
• Salesforce, ServiceNow, Confluence, Azure DevOps Wiki, etc.
Generative AI security and compliance
considerations
The large language models (LLM) Copilot Studio depends upon are the foundation models
trained by OpenAI.
Generative answers uses GPT-3.5 Turbo, but as new models outperform previous ones, they
can be updated.
Models are hosted on internal Azure OpenAI services, honoring the Microsoft Services Trust
boundary.
Models are accessed and used following Microsoft Responsible AI principles and policies.
Custom instructions can be added to generative answers to influence format and filters.
Data storage and processing for Copilot Studio generative AI capabilities may result in data
movement across regional boundaries. When local data storage and processing aren’t
available in the region, environment settings are available to disable features requiring Azure
OpenAI or Bing Search, preventing data movement.
Copilot Studio does not collect, nor provide any customer data, for use in the training of LLMs.
For operations purposes only, conversations are stored temporarily in a service-operated and
secured store so that authorized Microsoft employees via Secure Access Workstations (SAWs)
with Just-In-Time (JIT) access requests may respond to support requests. Access can be
controlled through Customer Lockbox.
Lab time!
Lab 05 [30]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Lab 06: Use generative AI
orchestration
to interact with your connectors
Demo Lab 6
Generative AI orchestration

Test and trace feature to see it in

action
Dynamically chains actions and
topics together to produce actionable
responses
Handle multi-intent queries that
were not anticipated or built by the
user
Automatically slot-fill further
details to get the information
needed for the task

Powered by the Azure OpenAI

Service with LangChain concepts

Public Preview
Lab time!
Lab 06 [20]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Lab 07: Invoke AI Builder prompts
Demo Lab 7
Lab time!
Lab 07 [25]
minutes

© Copyright Microsoft Corporation. All rights reserved.

Q&A
Resources
 Copilot Studio trial aka.ms/TryCopilotStudio

Copilot Studio website aka.ms/CopilotStudio

Resources Official Blog aka.ms/CopilotStudioBlog

Copilot Studio Demo aka.ms/CopilotStudioDemo

Product documentation aka.ms/CopilotStudioDocs

Follow-up Session on 1/31 Product guidance aka.ms/CopilotStudioGuidance

aka.ms
Implementation guide
/CopilotStudioImplementationGuid
e
Community aka.ms/CopilotStudioCommunity