Chapter: Case Study: Automa ng "Contoso's" Intelligent Approval Engine

Objec ve

In this final chapter of the Power Automate module, you will synthesize your knowledge of
Cloud Flows, AI Builder, and Advanced Approvals to build a real-world Intelligent Approval
Engine. You will explore how "Contoso," a global retail giant, transformed its chao c
procurement process into a streamlined, AI-first opera on. By the end of this chapter, you will
be able to design complex, mul -stage orchestra ons that balance automated speed with
human accountability.

The Business Challenge: Procurement Paralysis

"Contoso" manages over 5,000 purchase requests (PRs) monthly. Their legacy process was
purely manual:

 Inconsistency: Similar requests were approved by some managers and rejected by

others.

 Speed: The average approval me was 12 days, leading to missed vendor discounts.

 Visibility: Employees had no idea where their requests were in the "email black hole."

The Strategy: The "AI-First" Filter

The goal was to build an engine that could:

1. Extract data from vendor quotes automa cally.

2. Analyze the risk and sen ment of the request.

3. Route the request based on the dollar amount and AI-calculated risk score.

4. No fy stakeholders in real- me.

Core Concepts : The Orchestra on Layer

To solve Contoso's problem, the architecture relies on the "Orchestrator Pa ern." This involves
a single Parent flow that directs traffic between various services.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
1. The Mul -Model Pipeline

An intelligent engine o en requires more than one AI model. In Contoso's case:

 Document Processing Model: Extracts the "Total Amount" and "Line Items" from the
vendor PDF.

 Category Classifica on Model: Iden fies the type of spend (e.g., "IT Hardware" vs.
"Office Supplies") to determine which department pays.

2. The Logic Matrix

The flow uses Condi onals and Switches to handle rou ng:

 Low Value (< $500): If AI confidence is > 95%, the request is auto-approved.

 Medium Value ($500 - $5,000): Requires a single manager's approval.

 High Value (> $5,000): Requires a mul -stage parallel approval from both Finance and
the Department Head.

Prac cal Implementa on: Step-by-Step

This example demonstrates the "High-Value" path of the Contoso engine.

1. The Trigger & Extrac on The flow starts when a file is uploaded to the "Quote Drop"
SharePoint folder. The first ac on is Extract informa on from documents, using the "Contoso
Vendor Model."

2. The Risk Assessment A Parse JSON step takes the extracted line items and calculates the
total. A Condi on checks if the vendor is on the "Approved Vendor List" in Dataverse.

3. The Adap ve Card Construc on Instead of a standard email, the flow builds a Teams
Adap ve Card.

 AI Insight: "The AI has detected this quote is 15% higher than our previous order for the
same items."

 Ac on: Bu ons for "Approve," "Reject," and "Request More Info."

4. The Parallel Approval For requests over $5,000, the flow uses the Parallel Branch feature.

 Branch A: Sends an approval to the Finance Team.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Branch B: Sends an approval to the Department Lead. The flow waits for both branches
to complete before proceeding.

5. Final Closure Once approved, a Desktop Flow (RPA) is triggered to log into the legacy ERP
system and generate a Purchase Order (PO) number, which is then emailed back to the original
requester.

Knowledge Check: Mul ple Choice Ques ons

1. In the Contoso case study, why was a Parallel Branch used for high-value requests?

 A) To make the flow run twice as fast.

 B) To allow both the Finance Team and the Department Head to review the request
simultaneously, rather than wai ng for one to finish before star ng the other.

 C) To save money on AI Builder credits.

 D) Because SharePoint requires parallel branches for all PDF files.

Correct Answer: B.

Explana on: Parallel branches are essen al for reducing "Process Latency." By allowing two
par es to review at the same me, the overall approval cycle is shortened.

2. What role did 'Confidence Scores' play in Contoso's 'Auto-Approval' logic?

 A) They determined the font size of the email.

 B) They acted as a "Safety Valve," ensuring that only requests where the AI was highly
certain were approved without human eyes.

 C) They were used to calculate the vendor's discount.

 D) They are not used in Cloud Flows.

Correct Answer: B.

Explana on: In intelligent automa on, we never trust AI blindly. Se ng a threshold (e.g., 95%
confidence) ensures that the "Auto-Pilot" only engages when the data is high-fidelity.

3. Which component was used to bridge the modern Cloud Flow with Contoso's 'Legacy ERP'
system that had no API?

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  A) Power BI

 B) A Custom Page

 C) A Desktop Flow (RPA)

 D) An Adap ve Card

Correct Answer: C.

Explana on: RPA (Desktop Flows) is the industry-standard solu on for "The Last Mile" of
automa on—connec ng modern AI workflows to older systems that cannot communicate via
web services or APIs.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: From Chatbots to Agents: The Paradigm Shi in Copilot Studio

Objec ve

In this chapter, you will explore the evolu on of conversa onal AI from sta c, rule-based
Chatbots to autonomous, goal-oriented AI Agents. You will learn about the "Agen c" mindset,
where Copilots no longer just talk but act by u lizing tools and reasoning. By the end of this
chapter, you will be able to dis nguish between intent-based and genera ve conversa ons and
understand how to architect agents that can navigate complex business processes
independently.

Core Concepts

The shi from chatbots to agents represents a fundamental move from Linear Logic to Dynamic
Orchestra on. In the old world, you mapped every possible path; in the new world, you provide
the AI with resources and a goal.

1. The Tradi onal Chatbot (Intent-Based)

Tradi onal bots rely on a "Call and Response" architecture.

 Trigger Phrases: The bot waits for a specific phrase (e.g., "Check order status").

 Decision Trees: You must manually draw the flow of the conversa on.

 Limita ons: If a user asks something slightly outside the script, the bot fails (the "I didn't
understand that" loop).

2. The Modern AI Agent (Genera ve & Dynamic)

Agents u lize Large Language Models (LLMs) to understand context and intent without pre-
defined paths.

 Genera ve Answers: The agent "reads" your documents (Knowledge) to answer

ques ons in real- me.

 Dynamic Chaining: Instead of a fixed flow, the agent selects which "Ac on" (Plugin) to
call based on the user's request.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Reasoning: The agent can plan mul ple steps to solve a problem (e.g., "Check the
inventory, find the shipping date, and then email the customer").

3. The "Brain" of the Agent: The Orchestrator

In Copilot Studio, the Orchestrator is the logic layer that sits between the user and the available
tools. When a user asks a ques on, the Orchestrator:

1. Analyzes the intent.

2. Iden fies required data or ac ons.

3. Executes the necessary plugins.

4. Synthesizes a response based on the result.

Prac cal Case Study: The "Autonomous Logis cs Coordinator"

The Challenge: "Global-Freight" used a tradi onal bot that could only answer "Where is my
package?" if the user provided an exact tracking number. If the user asked "Why is my shipment
delayed?", the bot would fail.

The Solu on Design:

1. Knowledge Integra on: The agent is "grounded" in the company's delay policy
documents and real- me weather reports via SharePoint.

2. Agen c Ac ons: The agent is given a Plugin (Power Automate) that can look up the
"Shipment ID" based on the user's email address.

3. Outcome: A user asks: "My shipment is late, what's going on?"

o The agent iden fies the user, looks up their shipment, sees a weather delay in
the transit city, and explains: "Your shipment is delayed due to a storm in Delhi;
per our policy, we've upgraded you to priority shipping."

Step-by-Step Example: Enabling Genera ve AI and Knowledge

This example guides you through shi ing your Copilot from "Rule-Based" to "Genera ve."

1. Create a New Copilot:

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  In Copilot Studio, click New Copilot.

 Give it a name and a primary language.

2. Add Knowledge (The "Grounding" Step):

 Go to the Knowledge tab.

 Click Add Knowledge and select Public Website or SharePoint.

 Enter the URL of your company's FAQ or manual. The agent will now use this to answer
ques ons it wasn't specifically programmed for.

3. Enable Genera ve Answers:

 Go to Genera ve AI se ngs.

 Select the Genera ve op on. This allows the bot to use its "Knowledge" sources to
construct unique responses rather than just showing a link.

4. Test the Shi :

 Ask the bot a ques on that isn't in your "Topics" but is in your "Knowledge" documents.
The agent will analyze the document and provide a tailored answer, demonstra ng the
move away from fixed trigger phrases.

Knowledge Check: Mul ple Choice Ques ons

1. What is the primary difference between a tradi onal chatbot and an AI agent in Copilot
Studio?

 A) Chatbots use colors, while agents use text.

 B) Chatbots rely on pre-defined intent maps and trigger phrases, while agents use
genera ve AI and dynamic orchestra on to solve problems.

 C) Agents are only available on mobile phones.

 D) Chatbots are more intelligent than agents.

Correct Answer: B.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Explana on: The shi to "Agen c" behavior means moving away from rigid, manual
conversa on trees and toward a system where the AI uses reasoning to determine the best path
to reach a goal.

2. In an 'Agen c' workflow, what is the role of 'Knowledge'?

 A) It is a list of passwords.

 B) It acts as the "Grounding" data that the agent uses to provide accurate, company-
specific answers instead of general AI hallucina ons.

 C) It is a type of Power App.

 D) It prevents the bot from talking to users.

Correct Answer: B.

Explana on: Knowledge provides the context. Without it, an agent is just a general LLM; with it,
the agent becomes an expert on your specific business rules and data.

3. What happens during 'Dynamic Chaining' in Copilot Studio?

 A) The bot gets stuck in a loop.

 B) The agent automa cally selects and executes the appropriate plugins or ac ons
needed to fulfill a user's request without a manual "Go To" step.

 C) The bot sends an email to the developer.

 D) The conversa on is deleted.

Correct Answer: B.

Explana on: Dynamic Chaining is the hallmark of an agent. It allows the AI to "plan" its own
route to a solu on by picking the right tools (Ac ons) at the right me based on the
conversa on's flow.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Topic Management: Trigger Phrases and En ty Extrac on

Objec ve

In this chapter, you will master the art of conversa on design within Copilot Studio. You will
learn how to build precise Topics that serve as the building blocks of an agent's exper se.
Specifically, you will explore how to cra effec ve Trigger Phrases to ini ate conversa ons and
how to use En ty Extrac on to capture and "remember" cri cal informa on provided by the
user. By the end of this chapter, you will be able to build intelligent dialogue flows that
understand user intent and collect structured data for processing.

Core Concepts

Topic Management is the process of organizing an agent's "Brain." Each topic represents a
specific subject or capability the agent possesses.

1. Trigger Phrases: The "Star ng Gun"

A trigger phrase is a word or sentence that a user might say to start a specific topic.

 The "Goldilocks" Rule: You should provide 5–10 diverse phrases. Too few, and the agent
won't recognize the intent; too many overlapping phrases, and the agent will get
confused between topics.

 Varia on: Use different sentence structures (e.g., "I want to buy a laptop" vs. "Laptop
pricing").

 AI Interpreta on: Modern agents use Natural Language Understanding (NLU), so they
can trigger a topic even if the user's input isn't a 100% exact match to your phrase.

2. En ty Extrac on: Capturing Data

An En ty is a specific piece of informa on that the agent "extracts" from a user's message.

 Prebuilt En es: Copilot Studio comes with ready-made en es for common data like
Age, City, Color, Date and Time, and Organiza on.

 Custom En es: You can create your own lists (e.g., a list of your company’s product
models) or use Regular Expressions (Regex) for pa erns like employee IDs.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Slots and Variables: When an en ty is extracted, it is stored in a Variable. This allows the
agent to "remember" the detail (e.g., the user's name) and use it later in the
conversa on or pass it to a Power Automate flow.

3. Slot Filling

This is the process of an agent proac vely asking for missing informa on. If a topic requires a
"City" and a "Date" to provide a weather report, but the user only says "What's the weather in
Delhi?", the agent will recognize the City is filled and only ask for the Date.

Prac cal Case Study: The "Smart" Hotel Concierge

The Challenge: "Grand Stay Hotels" wants an agent to handle room service orders. The agent
needs to know the Item the guest wants and their Room Number.

The Solu on Design:

1. Topic: Created a topic named "Order Room Service."

2. Trigger Phrases: "I'm hungry," "I'd like to order food," "Room service menu."

3. En es:

o Used a Custom En ty called FoodItems (Pizza, Burger, Salad).

o Used the Prebuilt En ty Number for the RoomNumber.

4. The Interac on: A guest says, "I want a Burger sent to room 402."

o The agent extracts Burger into varFood and 402 into varRoom.

o Instead of asking "What would you like?" the agent skips straight to confirma on:
"Got it! One Burger is on its way to room 402."

Step-by-Step Example: Building a Lead Capture Topic

This example guides you through crea ng a topic that extracts a user's email address.

1. Create the Topic: Go to Topics -> Add a topic -> From blank. Name it "Lead Capture."

2. Add Trigger Phrases: Enter: "Contact sales," "I'm interested in a demo," "Speak to a
representa ve."

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
 3. Add a Ques on Node: Click the + icon and select Ask a ques on.

o Ques on: "What is your professional email address?"

o Iden fy: Select the Email prebuilt en ty.

o Save response as: Rename the variable to varUserEmail.

4. Add a Message Node: Click + and select Send a message.

o Message: "Thank you! We will send the details to {varUserEmail} shortly."

5. Test: In the test pane, type "I want a demo." When the agent asks for your email, type
"sachin@[Link]." Observe how the agent extracts the email and plays it back to
you.

Knowledge Check: Mul ple Choice Ques ons

1. What happens if a user provides an en ty (like their name) in their very first message
before the agent even asks for it?

 A) The agent crashes.

 B) The agent ignores the informa on and asks for the name anyway.

 C) The agent "extracts" the informa on and skips the ques on node related to that
en ty.

 D) The agent deletes the message.

Correct Answer: C.

Explana on: This is known as "Slot Filling." Intelligent agents are designed to recognize
informa on as soon as it is provided, making the conversa on feel more natural and efficient.

2. Which en ty type would you use if you wanted to capture a specific product code that
always follows the format of two le ers followed by four numbers (e.g., AB1234)?

 A) Prebuilt "Number" En ty

 B) Prebuilt "Organiza on" En ty

 C) Custom En ty with "Regular Expression" (Regex)

 D) Custom En ty with "List"

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-
applica ons/?referralCode=857964209E5CBDF3B947
Correct Answer: C.

Explana on: Regular Expressions are the best tool for capturing data that follows a strict,
predictable pa ern like ID numbers, license plates, or specialized product codes.

3. What is the recommended number of trigger phrases for a single topic to ensure the AI
understands the intent clearly?

 A) Exactly 1

 B) 5 to 10 diverse phrases

 C) At least 100 phrases

 D) Trigger phrases are not needed for agents.

Correct Answer: B.

Explana on: Providing 5 to 10 varia ons helps the NLU engine learn the "shape" of the intent.
Providing too many can lead to "Topic Overlap," where the agent can't dis nguish between two
similar topics.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Using Genera ve Answers for Dynamic Knowledge Retrieval

Objec ve

In this chapter, you will master the implementa on of Genera ve Answers in Copilot Studio.
You will learn how to move beyond sta c, hard-coded responses by connec ng your agent to
real- me data sources like websites, SharePoint, and internal files. By the end of this chapter,
you will be able to configure "Knowledge-Base Grounding," understand how the AI synthesizes
informa on from mul ple documents, and implement safety filters to ensure accurate and
professional outputs.

Core Concepts

Genera ve Answers transform an agent from a scripted bot into a Knowledge Expert. Instead of
a developer wri ng every word, the agent uses an LLM to "read" your sources and "write" a
response on the fly.

1. The RAG Pa ern (Retrieval-Augmented Genera on)

Genera ve Answers u lize a technical architecture called RAG.

 Retrieval: When a user asks a ques on, the agent searches your connected knowledge
sources (SharePoint, Web, etc.) for relevant snippets.

 Augmenta on: It adds these snippets to the prompt it sends to the AI.

 Genera on: The AI writes a natural-sounding answer based only on the retrieved
snippets.

2. Knowledge Sources and Grounding

"Grounding" is the process of restric ng the AI's knowledge to your specific data to prevent
"hallucina ons" (making things up).

 Public Websites: The agent can crawl public FAQs or documenta on pages.

 SharePoint/OneDrive: The agent can access internal Word, PDF, and PowerPoint files
(respec ng the user's exis ng permissions).

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Uploaded Files: You can upload specific policy documents directly into Copilot Studio
(up to 20MB per file).

3. Content Modera on & Boosted Conversa ons

You can control the "Crea vity" and "Safety" of the responses using the Content Modera on
se ng:

 High: Strict answers only from the text; highly accurate but less conversa onal.

 Medium: (Default) Balanced approach.

 Low: More crea ve and flexible, but higher risk of including informa on outside the
source.

Prac cal Case Study: The "Live" HR Policy Assistant

The Challenge: "Tech-Global" has a 200-page HR handbook that is updated every month.
Employees are frustrated because the old chatbot o en gives outdated informa on because the
developers forgot to update the bot's sta c topics.

The Solu on Design:

1. Direct Connec on: The HR agent is connected to the SharePoint folder where the live
PDF handbook is stored.

2. Dynamic Retrieval: When an employee asks, "What is the new remote work policy for
2026?", the agent doesn't look for a trigger phrase.

3. Synthesis: It instantly "reads" the latest version of the PDF, finds the 2026 sec on, and
summarizes the rules.

4. Cita ons: The agent provides a link to the specific page in the PDF so the employee can
verify the facts.

Step-by-Step Example: Connec ng a Website as Knowledge

This example demonstrates how to make your agent an expert on a specific website in under
two minutes.

1. Access Knowledge Se ngs:

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  In Copilot Studio, open your agent and go to the Knowledge tab.

2. Add a Website Source:

 Click + Add Knowledge.

 Select Public Website.

 Enter a URL (e.g., [h ps://[Link] .com/en-us/pmi-cpmai-

guide](h ps://[Link] .com/en-us/pmi-cpmai-guide)).

 Click Add.

3. Configure Genera ve Se ngs:

 Go to Se ngs -> Genera ve AI.

 Ensure "Use genera ve answers" is enabled.

 Set Content Modera on to "High" to ensure the agent only uses the website's text.

4. Test the Real- me Knowledge:

 In the Test pane, ask a specific ques on found on that website (e.g., "What are the
domains for the CPMAI exam?").

 Observe how the agent provides a summarized answer and includes a [1] cita on link
back to the source website.

Knowledge Check: Mul ple Choice Ques ons

1. What is the primary purpose of 'Cita ons' in Genera ve Answers?

 A) To make the bot look smarter.

 B) To provide transparency by showing the user exactly which source document or

website was used to generate the answer.

 C) To increase the cost of the conversa on.

 D) To prevent the bot from answering ques ons.

Correct Answer: B.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Explana on: Cita ons are a cri cal trust mechanism. They allow users to verify the AI's
response against the original "grounding" document, which is essen al for business and legal
accuracy.

2. If an agent is connected to a SharePoint site, what determines which files the agent can
'read' to answer a specific user's ques on?

 A) The agent can read every file in the company.

 B) The agent respects the user's exis ng SharePoint permissions; it will only find
informa on the user already has the right to see.

 C) The agent only reads files that are colored blue.

 D) The developer must manually grant permission for every file.

Correct Answer: B.

Explana on: This is a core security feature of Copilot. It ensures that sensi ve data (like payroll)
isn't leaked to users who don't have permission to view those specific folders or files.

3. Which 'Content Modera on' level would you choose if you are building a medical or legal
agent where accuracy is more important than being conversa onal?

 A) Low

 B) Medium

 C) High

 D) Off

Correct Answer: C.

Explana on: The 'High' se ng forces the LLM to stay extremely close to the provided text.
While it might sound less "friendly," it significantly reduces the chance of the AI adding incorrect
external informa on.

Expert Guide: How would you like to explore the integra on of custom AI models into these
conversa ons next?

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Integra ng Dataverse and Power Automate into Topics

Objec ve

In this chapter, you will learn how to turn a conversa onal agent into an Ac on-Oriented Agent.
You will explore the technical bridge between Copilot Studio, Dataverse, and Power Automate.
By the end of this chapter, you will be able to trigger background workflows from a
conversa on, retrieve real- me data from Dataverse to answer user queries, and send
structured updates back to your database—all without the user leaving the chat interface.

Core Concepts

Integra ng external logic is what separates a "FAQ Bot" from a "Business Agent." This
integra on relies on the exchange of variables between the chat interface and the Power
Pla orm backend.

1. The "Call an Ac on" Workflow

In Copilot Studio, you don't write code to talk to a database. Instead, you "Call an Ac on."

 The Input: The agent passes variables collected during the chat (e.g., varEmail,
varOrderNumber) to a Power Automate flow.

 The Processing: The flow performs the "heavy li ing," such as searching Dataverse,
calcula ng a discount, or calling an external API.

 The Output: The flow sends the result back to the agent as a new variable (e.g.,
flow_OrderStatus), which the agent then displays to the user.

2. Dataverse as the "Single Source of Truth"

By connec ng to Dataverse, your agent gains "Long-term Memory."

 Read Opera ons: The agent can look up a user’s loyalty points or pending cases based
on their email.

 Write Opera ons: The agent can create new lead records, update contact info, or log
support ckets directly into the system.

3. Power Automate: The Orchestrator

Power Automate acts as the "hands" of the agent. While the agent "talks," the flow "acts."

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Authen ca on: Flows handle the security context, ensuring that the agent can only
access data it is permi ed to see.

 Forma ng: Flows can take raw data from Dataverse and "clean it up" (e.g., forma ng a
date or currency) before the agent speaks it to the user.

Prac cal Case Study: The "Instant" Refund Checker

The Challenge: "Swi -Retail" customer support is overwhelmed by users asking "Where is my
refund?" Agents spend 5 minutes per call looking up the transac on in the database.

The Solu on Design:

1. Conversa on: The agent asks the user for their Transac on ID.

2. Logic: The agent calls a Power Automate Flow, passing the ID.

3. Ac on: The flow searches the Dataverse Sales_Transac ons table.

4. Retrieval: The flow finds the "Refund Status" and "Es mated Date."

5. Response: The flow sends these two pieces of data back to the agent.

6. Outcome: The agent tells the user: "Your refund was processed on May 2nd and should
arrive by May 7th." The en re interac on takes 15 seconds.

Step-by-Step Example: Crea ng a "Lead Creator" Ac on

This example demonstrates how to take a user's name and email from a chat and save it into a
Dataverse table.

1. Collect User Data (Copilot Studio):

 Create a Topic "Sign Up for Newsle er."

 Use Ques on Nodes to capture varUserName and varUserEmail.

2. Create the Flow:

 Click + Add Node -> Call an Ac on -> Create a flow.

 Trigger: "Power Virtual Agents" (legacy name, used for compa bility).

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Input: Add two text inputs: Name and Email.

3. Connect to Dataverse:

 Add a step: Add a new row (Dataverse).

 Table: Leads.

 Name Field: Map to the Name input from the flow trigger.

 Email Field: Map to the Email input.

4. Return Success to Agent:

 In the final flow step (Return value(s) to Power Virtual Agents), add a Text output called
Status.

 Set the value to: "Success! You have been added to our database."

5. Finalize in Copilot Studio:

 Back in the topic, map your chat variables to the flow inputs.

 Add a Message Node to display the {Status} variable returned by the flow.

Knowledge Check: Mul ple Choice Ques ons

1. When a Power Automate flow is called from Copilot Studio, what must be the very last step
in the flow for the agent to receive informa on back?

 A) Terminate

 B) Respond to a Power App or flow (or Return values to PVA)

 C) Send an email

 D) Update a row in Dataverse

Correct Answer: B.

Explana on: Without the "Return/Respond" ac on, the flow will complete its task, but the
agent will never receive the data it needs to finish the conversa on with the user.

2. Why is it be er to use a Power Automate flow to query Dataverse instead of having the
agent try to "read" the database directly?

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  A) Flows are more colorful than agents.

 B) Flows allow for complex logic, data forma ng, and mul -step security checks before
the data is presented to the user.

 C) Agents cannot connect to Dataverse.

 D) Flows make the agent run on a desktop.

Correct Answer: B.

Explana on: Power Automate provides a robust "Middleware" layer. It can filter, sort, and
transform raw database data into "human-friendly" text that the agent can easily speak.

3. In the context of integra ng an agent with Dataverse, what is an 'Input Variable'?

 A) A piece of data the agent sends TO the flow (e.g., the User's ID).

 B) A piece of data the flow sends back TO the agent.

 C) A password for the database.

 D) A type of Excel spreadsheet.

Correct Answer: A.

Explana on: Input variables are the parameters the agent "feeds" into the flow so the flow
knows which specific record to look for or update.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Managing Variables and Contextual Persistence

Objec ve

In this chapter, you will master the "Memory" of your AI Agent. You will learn how to use
Variables to store user informa on, maintain Contextual Persistence across different topics,
and manage variable Scopes (Topic vs. Global). By the end of this chapter, you will be able to
design sophis cated agents that remember a user's name, preferences, and previous inputs,
crea ng a personalized and seamless conversa on flow.

Core Concepts

Variables are containers that hold data. In Copilot Studio, they allow the agent to move from a
"Stateless" interac on (forge ng everything immediately) to a "Stateful" one (remembering
details).

1. Variable Scopes

Understanding where a variable "lives" is crucial for preven ng data leaks and ensuring logical
flow:

 Topic Scope: The variable is only accessible within the specific topic where it was
created. It is deleted once the topic ends. Use this for temporary data like a single
calcula on.

 Global (Bot) Scope: The variable is accessible across all topics during a single session.
Use this for data like UserName or UserEmail.

 System Variables: Built-in variables provided by Copilot Studio, such as

[Link] or Conversa [Link].

2. Contextual Persistence

Persistence refers to the agent's ability to "carry" informa on from one part of the conversa on
to another.

 Implicit Context: The agent uses Natural Language Understanding (NLU) to maintain the
"vibe" of the chat.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Explicit Context: The developer manually saves a value (e.g., varSelectedProduct) so that
when the user switches topics from "Product Info" to "Pricing," the agent already knows
which product to talk about.

3. Variable Types & Ini aliza on

Variables are "Strongly Typed" in Copilot Studio. Common types include:

 String: Text data.

 Number: Numeric data for calcula ons.

 Boolean: True/False values.

 Choice/Op on Set: Specific values from a list.

Prac cal Case Study: The "Returning Customer" Experience

The Challenge: "Quick-Print" customers o en ask about "Pricing" a er they have already spent
10 minutes configuring a "Business Card" design. In the old system, the agent would ask "What
product are you interested in?" again, frustra ng the customer.

The Solu on Design:

1. Global Variable: When the user selects "Business Cards" in the Configura on topic, the
agent saves it to a Global Variable called [Link].

2. Persistence: When the user triggers the Pricing topic, the agent first checks if
[Link] has a value.

3. Intelligent Response: Instead of asking for the product, the agent says: "Based on the
Business Cards you just designed, the total price is $25."

4. Outcome: The "Customer Effort Score" improved significantly because the agent felt
"smarter" and more a en ve.

Step-by-Step Example: Se ng and Using a Global Variable

This example demonstrates how to capture a user's name in one topic and use it to greet them
in a completely different topic.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
1. Create the Capture Topic (Topic A):

 Add a Ques on Node: "May I know your name?"

 Iden fy: Person Name.

 Save response as: Click the variable, change the name to [Link], and change
the Scope to Global (any topic can access).

2. Create the Gree ng Topic (Topic B):

 Create a new topic triggered by the phrase "Help me."

 Add a Message Node.

 Message: "Hello {[Link]}, how can I assist you today?"

3. Test the Persistence:

 In the test pane, start Topic A and provide your name.

 Immediately type "Help me" (triggering Topic B).

 Observe how the agent uses your name without asking for it again.

Knowledge Check: Mul ple Choice Ques ons

1. Which variable scope should you use if you want the agent to remember the user's
'Department' throughout the en re chat session, even as they switch between HR and IT
topics?

 A) Topic Scope

 B) Global (Bot) Scope

 C) System Scope

 D) Hidden Scope

Correct Answer: B.

Explana on: Global variables are designed for "Session Persistence." They ensure that once a
piece of data is captured, it is available to every topic the agent might trigger un l the session
ends.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
2. What happens to a variable with a 'Topic' scope once the conversa on moves to a different
topic?

 A) It is saved to Dataverse automa cally.

 B) It is cleared from memory and is no longer accessible.

 C) It becomes a Global variable.

 D) It is emailed to the admin.

Correct Answer: B.

Explana on: Topic-scoped variables are temporary. They are used for "Local Logic" to keep the
agent's memory clean and prevent variables from conflic ng with each other across different
parts of the bot.

3. Why is it important to select the correct 'En ty' (like Email or Date) when saving a variable?

 A) It makes the variable icon look be er.

 B) It ensures the variable is "Typed" correctly, which prevents errors when passing that
data to Power Automate or using it in calcula ons.

 C) It is the only way to save a variable.

 D) It allows the bot to work in mul ple languages.

Correct Answer: B.

Explana on: Copilot Studio uses "Strong Typing." If you save a "Number" as a "String," you
won't be able to perform mathema cal opera ons on it later. Matching the En ty to the
Variable Type is a technical best prac ce for stable agents.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Configuring Mul -Turn Conversa ons and Slot Filling

Objec ve

In this chapter, you will master the mechanics of Mul -Turn Conversa ons, the founda on of
natural human-AI interac on. You will learn how to design dialogue flows that span mul ple
exchanges and how to implement Slot Filling to collect missing informa on efficiently. By the
end of this chapter, you will be able to build agents that feel intui ve, remember context within
a session, and proac vely ask for necessary data without repea ng themselves.

Core Concepts

A "Mul -Turn" conversa on is any dialogue where the agent and user exchange informa on
over several rounds to complete a task. This is the opposite of a "Single-Turn" interac on (like a
simple FAQ lookup).

1. The Logic of Mul -Turn Dialogue

In a mul -turn scenario, the agent must maintain Contextual State. It needs to know what has
already been said so it can ask the next logical ques on.

 State Management: The agent uses variables to "keep track" of the conversa on's
progress.

 Topic Switching: A mul -turn agent can handle a user "pivo ng" (e.g., the user is in the
middle of a flight booking but suddenly asks about baggage rules) and then return to the
original task.

2. Slot Filling: The "Checklist" Approach

Slot Filling is an AI technique where the agent iden fies a set of required parameters (slots)
needed to execute an ac on.

 The Proac ve Agent: If a user says, "I want to book a flight to Delhi," the agent iden fies
that the Des na on slot is filled but the Date and Departure City slots are empty.

 Dynamic Ques oning: The agent skips the ques on about the des na on and only asks
for the missing pieces.

3. Handling Interrup ons (Digressions)

High-quality mul -turn conversa ons allow for Digressions.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Example: While filling out a complex insurance form, a user might ask, "Wait, what does
'deduc ble' mean?"

 The Agent's Response: The agent should trigger a "Defini on" topic, answer the
ques on, and then use the Go back to previous topic logic to resume the form exactly
where the user le off.

Prac cal Case Study: The "Smart" Pizza Bot

The Challenge: "Mama’s Pizza" has a complex ordering process. They need to know the Size,
Crust Type, and Toppings. Tradi onal bots ask these one by one, even if the user provides the
informa on upfront.

The Solu on Design:

1. En ty Mapping: Define en es for PizzaSize (Small, Large) and Crust (Thin, Pan).

2. Ini al Input: A user says, "I'd like a large thin-crust pizza."

3. Slot Filling Execu on:

o Size slot: Large (Filled)

o Crust slot: Thin-crust (Filled)

o Toppings slot: Empty

4. Mul -Turn Result: The agent bypasses the first two ques ons and immediately asks,
"Great! And what toppings would you like on your large thin-crust pizza?"

5. Outcome: The conversa on feels human-like, reduces "Interac on Fric on," and speeds
up the ordering process.

Step-by-Step Example: Implemen ng Slot Filling for an IT Ticket

This example demonstrates how to configure a topic to intelligently collect cket details.

1. Set up the Topic:

 Create a topic: "Create IT Support Ticket."

 Trigger Phrases: "I have a technical issue," "My laptop is broken," "Printer error."

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
2. Add Ques on Nodes with "Iden fy":

 Ques on 1: "Which device is having the issue?"

o Iden fy: Create a Custom En ty DeviceTypes (Laptop, Desktop, Printer).

o Save as: varDevice.

 Ques on 2: "Please describe the problem."

o Iden fy: User's en re response.

o Save as: varDescrip on.

3. Configure "Skip if Value Exists":

 Click the three dots (...) on each Ques on node.

 Ensure the Ques on behavior is set to Skip if value exists. This is the toggle that enables
Slot Filling.

4. The Final Ac on:

 Add a Message Node: "Got it. I'm opening a cket for your {varDevice} regarding:
{varDescrip on}."

5. Test the Flow:

 Test 1: Type "My printer is broken." The agent should skip the "Which device?" ques on
and ask for the descrip on.

 Test 2: Type "I have a technical issue." The agent should ask for both.

Knowledge Check: Mul ple Choice Ques ons

1. What is the primary benefit of 'Slot Filling' in a conversa on?

 A) It makes the agent talk more.

 B) It allows the agent to skip redundant ques ons if the user has already provided the
informa on in a previous message.

 C) It deletes the user's data a er 24 hours.

 D) It requires the user to type in a specific code.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Correct Answer: B.

Explana on: Slot filling is the hallmark of an efficient agent. It ensures that the conversa on is
as brief and helpful as possible by only asking for missing informa on.

2. A user is in the middle of a 'Return Item' topic but suddenly asks, 'What are your store
hours?' What feature allows the agent to handle this and then return to the return process?

 A) Slot Filling

 B) Topic Digression

 C) Variable Persistence

 D) Global Trigger

Correct Answer: B.

Explana on: Digressions allow an agent to pause the current "Turn," address a side-request,
and then resume the original logic path without losing the collected "Slots."

3. If you want an agent to 'remember' a piece of data across a mul -turn conversa on that
spans three different topics, what scope should you give the variable?

 A) Topic Scope

 B) System Scope

 C) Global (Bot) Scope

 D) Permanent Scope

Correct Answer: C.

Explana on: Global variables stay in the agent's "Memory" for the dura on of the en re
session, allowing data to persist as the user moves from one topic to another.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Publishing Agents to Mul ple Channels (Teams, Web, Custom)

Objec ve

In this chapter, you will learn the final, cri cal stage of the agent lifecycle: Deployment. You will
explore how to transi on your agent from the "Test Canvas" to the real world. By the end of this
chapter, you will be able to publish your agent to Microso Teams, embed it into a Public
Website, and understand the technical requirements for connec ng to Custom Channels using
the Direct Line API.

Core Concepts

Publishing is the process of "Freezing" a version of your agent and making it accessible to users.
Un l you click Publish, any changes you make in the designer are invisible to your audience.

1. The Publica on Lifecycle

 Dra ing: Making changes in the Copilot Studio designer.

 Publishing: Crea ng a "Snapshot" of the current logic and knowledge.

 Channel Configura on: Deciding where that snapshot will appear (e.g., Teams, Mobile
App).

2. Standard Channels

Microso provides "Na ve" connectors for the most common business pla orms:

 Microso Teams: Ideal for internal employees. It supports Single Sign-On (SSO) and can
be pinned to the Teams sidebar.

 Demo Website: A temporary, password-protected site provided by Microso to show

the agent to stakeholders before a full rollout.

 Custom Website: Provides an <iframe> code or a JavaScript snippet to embed the agent
into your company’s exis ng portal.

3. Custom Channels (Direct Line)

For high-end intelligent applica ons, you may need to deploy the agent to a custom mobile app
or a specific hardware device (like a kiosk).

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Direct Line API: The technical "pipe" that allows your custom code to talk to the Copilot
Studio engine.

 Token Security: Custom channels require secret keys and tokens to ensure only
authorized applica ons can trigger your agent.

Prac cal Case Study: The "Everywhere" Support Agent

The Challenge: "City-U lity Co" wants their new AI agent to be accessible to three different
groups:

1. Field Workers: Using Microso Teams on their mobile phones.

2. Customers: Visi ng the public "Pay My Bill" website.

3. Developers: Accessing a custom IoT dashboard.

The Solu on Design:

 Teams Integra on: The agent is published to the Teams Channel, allowing workers to
use GPS-integrated topics.

 Web Embedding: The developer copies the HTML Snippet from the "Custom Website"
channel and pastes it into the footer of the u lity’s public site.

 Direct Line: The IoT team uses the Secret Key from the "Mobile App" channel to
integrate the agent's chat window directly into their proprietary hardware dashboard.

 Outcome: One single "Brain" (the agent) serves three completely different audiences
across three different pla orms, ensuring consistent informa on everywhere.

Step-by-Step Example: Embedding an Agent on a Web Page

This example demonstrates how to move your agent from Copilot Studio to a live webpage.

1. Publish the Agent:

 In Copilot Studio, click the Publish icon in the side menu.

 Click the Publish bu on. Wait for the success message.

2. Configure the Channel:

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Go to Se ngs -> Channels.

 Select Custom Website.

 Copy the provided HTML <iframe> code.

3. Deploy the Code:

 Open your website's HTML editor.

 Paste the code into the body of the page:

HTML

<iframe src="h ps://[Link] .com/webchat/bots/YOUR-BOT-ID"

frameborder="0"

style="width: 100%; height: 100%;">

</iframe>

4. Verify Connec vity:

 Load your website. The agent window should appear.

 Ask a ques on to ensure it can s ll access its Knowledge sources and Power Automate
flows.

Knowledge Check: Mul ple Choice Ques ons

1. A er making changes to a Topic in Copilot Studio, what MUST a developer do before those
changes become visible to users on Microso Teams?

 A) Restart Microso Teams.

 B) Click the 'Publish' bu on in Copilot Studio.

 C) Delete the agent and recreate it.

 D) Send an email to Microso support.

Correct Answer: B.

Explana on: Publishing is a manual gate. It ensures that you can work on "Dra s" and test them
internally without disrup ng the live experience for your users.
h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-
applica ons/?referralCode=857964209E5CBDF3B947
2. Which channel should you choose if you want to integrate your agent into a proprietary
mobile app that your company built from scratch?

 A) Demo Website

 B) Microso Teams

 C) Mobile App (Direct Line)

 D) Facebook Messenger

Correct Answer: C.

Explana on: The 'Mobile App' channel provides the API keys necessary for developers to "wire"
the agent into custom-built so ware environments using the Direct Line protocol.

3. What is the benefit of the 'Demo Website' channel?

 A) It is the final version of your website.

 B) It provides a quick, hosted URL that you can share with stakeholders to gather
feedback on the agent's performance before going live.

 C) It makes the agent run faster.

 D) It automa cally writes the code for your website.

Correct Answer: B.

Explana on: The Demo Website is a staging environment. It allows non-technical stakeholders
to "chat" with the agent in a realis c web environment to ensure the tone and accuracy are
correct before the official launch.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Authen ca on and Single Sign-On (SSO) for Agents

Objec ve

In this chapter, you will master the security protocols required to build enterprise-grade AI
agents. You will learn how to transi on from "Anonymous" bots to Authen cated Agents that
know exactly who they are talking to. By the end of this chapter, you will be able to configure
Single Sign-On (SSO), understand the role of Microso Entra ID (formerly Azure AD), and
implement secure data retrieval where the agent accesses user-specific records without asking
for a password.

Core Concepts

Authen ca on is the "ID Check" of the digital world. In Copilot Studio, it serves two purposes:
iden fying the user and authorizing the agent to act on their behalf.

1. Anonymous vs. Authen cated Access

 Anonymous (No Authen ca on): Anyone with the link can chat. The agent has no
access to private data.

 Only for Teams and Power Apps: Uses the user's exis ng login. Ideal for internal
corporate bots.

 Manual (OAuth2): Required for custom websites. The agent prompts the user to log in
via a "Sign In" card.

2. Microso Entra ID (Azure AD)

For intelligent applica ons within the Microso ecosystem, Entra ID is the "Iden ty Provider."

 Scopes: These are specific permissions (e.g., [Link], [Link]) that you grant to the
agent.

 App Registra on: To enable SSO, you must register the agent as an "App" in the Azure
Portal, crea ng a unique Client ID and Secret.

3. The Power of Single Sign-On (SSO)

SSO allows a user who is already logged into their computer or Teams to talk to the agent
without seeing a login prompt.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  The Seamless Experience: The agent silently retrieves the user's "Token."

 Contextual Data: Because the agent knows the user's iden ty, it can automa cally
access their specific Dataverse records, Outlook calendar, or OneDrive files using the
{[Link]} or {[Link]} system variables.

Prac cal Case Study: The "Personalized" IT Desk

The Challenge: "Finance-Corp" built an IT support agent. However, every me a user asked
"What is the status of my cket?", the agent had to ask for their Email and Ticket ID. Users
found this tedious and insecure.

The Solu on Design:

1. Iden ty Integra on: The developer enabled SSO using the company's Entra ID.

2. Silent Authen ca on: When a user opens the agent in Teams, the agent automa cally
knows they are "Sarah from Accoun ng."

3. Authen cated Query: Sarah asks: "Status of my cket?"

4. Ac on: The agent passes Sarah’s authen cated email to a Power Automate flow. The
flow queries Dataverse for only the ckets assigned to that specific email.

5. Outcome: The agent responds: "Hi Sarah, your laptop repair (Ticket #402) is currently in
'In Progress' status." No manual iden fica on was required.

Step-by-Step Example: Configuring Authen ca on for Teams

This example demonstrates how to secure an agent so it can recognize internal employees.

1. Enable Authen ca on in Copilot Studio:

 Go to Se ngs -> Security -> Authen ca on.

 Select Only for Teams and Power Apps. (This is the simplest form of SSO).

 Click Save.

2. Use System Variables:

 Create a new Topic.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Add a Message Node.

 Type: "Hello, I see you are logged in as {[Link]}

({[Link]})."

 Note: These variables are automa cally populated once authen ca on is enabled.

3. Test the Security:

 Open the Test pane. You should see a message sta ng that you are now authen cated.

 Try calling a Power Automate flow and no ce how you can now pass the [Link]
variable as a "hardened" iden ty marker that the user cannot spoof.

Knowledge Check: Mul ple Choice Ques ons

1. What is the primary benefit of enabling Single Sign-On (SSO) for an internal agent?

 A) It makes the agent's icons larger.

 B) It allows the user to interact with the agent without repeatedly entering their
username and password, as the agent leverages their exis ng login session.

 C) It allows the agent to talk to people outside the company.

 D) It reduces the amount of storage used by Dataverse.

Correct Answer: B.

Explana on: SSO removes "Fric on." By recognizing the exis ng secure session (like a Teams
login), the agent provides a smoother user experience while maintaining high security.

2. If an agent needs to read a user's Outlook Calendar, which security configura on is

required?

 A) Anonymous Access

 B) Authen cated Access with specific 'Scopes' (permissions) granted in Microso Entra
ID.

 C) A public website link.

 D) No configura on is needed; AI can read all data by default.

Correct Answer: B.
h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-
applica ons/?referralCode=857964209E5CBDF3B947
Explana on: Accessing private user data requires "Delegated Permissions." The user (or an
admin) must grant the agent the scope to read the calendar via the authen ca on provider.

3. Which system variable would you use to pass a user's unique iden fica on to a Power
Automate flow to ensure the flow only retrieves that user's specific data?

 A) {varUser}

 B) {[Link]}

 C) {[Link]}

 D) {Conversa [Link]}

Correct Answer: B.

Explana on: {[Link]} is a verified variable. When authen ca on is on, this is a

reliable way to filter database queries so users only see their own informa on.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Chapter: Case Study: Deploying a Mul -Language Customer Service Agent

Objec ve

In this concluding chapter of the Copilot Studio module, you will analyze the end-to-end
deployment of a Global AI Agent. You will explore how to manage a single agent that supports
mul ple languages, localized knowledge bases, and regional business rules. By the end of this
chapter, you will be able to configure Mul -Language Support, implement language-specific
branching, and understand the best prac ces for deploying a unified agent across a global
enterprise.

The Business Challenge: The Tower of Babel

"Omni-Retail Corp" operates in 15 countries across Europe, Asia, and the Americas. Previously,
they had 15 different chatbots—one for each country. This led to:

 Maintenance Nightmare: Any policy change (e.g., a new return window) had to be
updated 15 mes.

 Inconsistent Quality: The English bot was highly advanced, while the Italian and
Japanese bots were basic and prone to errors.

 Resource Drain: They needed 15 different developers to manage the separate logic
paths.

The Strategy: The "Single-Brain" Global Agent

Instead of fragmented bots, Omni-Retail moved to a Unified Mul -Language Agent. This agent
detects the user's language and dynamically switches its personality and knowledge retrieval
while keeping the core business logic (like Power Automate flows) iden cal.

Core Concepts : Interna onaliza on (i18n)

Designing for a global audience requires moving beyond simple transla on to Localiza on.

1. Language Detec on & User Preference

Copilot Studio can handle languages in two ways:

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
  Automa c Detec on: The agent analyzes the first few words of the user (e.g., "Hola" vs.
"Hello") and switches accordingly.

 Browser Se ngs: The agent reads the user's browser language header to set the ini al
tone.

2. Mul -Language Knowledge Base

For Genera ve Answers, you can connect the agent to localized data sources.

 If the user speaks Spanish, the agent queries the SharePoint/ES/Policies folder.

 If the user speaks English, it queries SharePoint/EN/Policies.

3. Transla on vs. Na ve Topics

 Machine Transla on: The agent uses an LLM to translate its standard responses on the
fly.

 Na ve Topics: For cri cal regions, developers can provide manually curated transla ons
for specific "Message Nodes" to ensure cultural nuance and legal accuracy.

Prac cal Implementa on: The Omni-Retail Architecture

1. Enabling Mul -Language Support The developer goes to Se ngs -> Languages and adds
"Primary" and "Secondary" languages. This creates a localiza on table within the bot.

2. The Dynamic Rou ng Logic Inside the "Check Order" topic, the agent uses a Condi on based
on the system variable [Link].

 If English: Use "Warm & Professional" tone.

 If Japanese: Use "Highly Formal & Respec ul" tone (Honorifics).

3. Unified Power Automate Integra on The agent calls the same "Get-Order-Status" flow
regardless of language. The flow retrieves raw data from Dataverse. The agent then translates
that raw data into the user's language.

4. Localized Hand-off If the AI cannot solve the problem, it iden fies the user's region and
routes the chat to a live agent in the correct regional call center (e.g., rou ng a French speaker
to the Paris office).

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Step-by-Step Example: Adding a Second Language

1. Configure Languages: Go to Se ngs -> Languages. Add "French" as a secondary

language.

2. Download Localiza on File: Download the .resx or Excel file containing all your bot's
current strings.

3. Translate & Upload: Provide the French transla ons for your messages and trigger
phrases in the file, then upload it back to Copilot Studio.

4. Test: In the test pane, click the Se ngs icon and change your language to French. Type
"Bonjour." The agent should now respond using the French strings you uploaded.

Knowledge Check: Mul ple Choice Ques ons

1. What is the primary advantage of using one Mul -Language Agent instead of 10 separate
local bots?

 A) It is cheaper to host.

 B) It ensures a "Single Source of Truth" where core business logic and Power Automate
flows only need to be built and maintained once.

 C) It makes the agent's icons change color based on the flag of the country.

 D) It prevents users from asking ques ons in English.

Correct Answer: B.

Explana on: Centralized management is the biggest win. You can update a "Refund Policy" flow
once, and it instantly applies to every language, reducing the risk of regional inconsistencies.

2. Which system variable is used to determine the user's current language se ng?

 A) {[Link]}

 B) {[Link]}

 C) {[Link]}

 D) {Conversa [Link]}

Correct Answer: B.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947
Explana on: {[Link]} stores the ISO language code (e.g., 'en-us', 'fr-fr'), allowing you to
create condi onal logic branches for localized responses.

3. When using 'Genera ve Answers' with a mul -language agent, what is the best prac ce for
Knowledge sources?

 A) Mix all languages into one folder.

 B) Translate all documents into English only.

 C) Use separate SharePoint folders or websites for each language to ensure the AI
retrieves the most culturally and legally accurate informa on for that specific region.

 D) Delete all documents and let the AI guess.

Correct Answer: C.

Explana on: Grounding the AI in localized documents prevents "Cross-transla on errors" and
ensures that the agent provides answers that comply with local regional laws and policies.

h ps://[Link]/course/microso -ab-410-cer fica on-intelligent-

applica ons/?referralCode=857964209E5CBDF3B947