This chapter covers Microsoft Copilot Studio, a powerful low-code tool for building custom copilots that integrate with Microsoft 365 and other data sources. For the AI-900 exam, this topic appears in the Generative AI domain and is tested on about 5-7% of questions, focusing on its purpose, key capabilities, and how it differs from pre-built copilots like Microsoft Copilot for Microsoft 365. You will learn the core components, the authoring experience, and the deployment options, all essential for answering exam questions correctly.
Jump to a section
Imagine you own a chain of hardware stores and want to create a custom assistant for each department. Instead of hiring a software developer to write code from scratch, you use a 'Bot Builder Workshop' that provides pre-built modules: a greeting module, a product lookup module, a checkout module, and a troubleshooting module. You can drag and drop these modules onto a canvas, connect them in a sequence, and configure each one with your specific product catalog and store policies. The workshop also has a 'knowledge base' where you can upload your store manuals and FAQs. When a customer asks a question, the workshop's AI engine first checks if it can answer using a pre-built module (like 'hours of operation'). If not, it searches the knowledge base and generates a response using natural language. If it still cannot answer, the workshop escalates to a human employee via a live chat interface. The workshop also logs all conversations so you can see which questions are frequently missed and update your knowledge base. This is exactly how Microsoft Copilot Studio works: it is a low-code platform to build custom copilots (conversational AI agents) by combining pre-built topics, AI-generated responses using generative AI, and human escalation, all without writing code from scratch.
What is Microsoft Copilot Studio?
Microsoft Copilot Studio (formerly Power Virtual Agents) is a low-code, drag-and-drop platform for creating custom conversational AI agents, called copilots. These copilots can be deployed on websites, mobile apps, Microsoft Teams, and other channels. The platform combines traditional bot-building (topics, entities, dialogs) with generative AI capabilities powered by Azure OpenAI Service. This means you can create copilots that not only follow predefined conversation flows but also generate natural language responses dynamically based on your own data sources.
Why Microsoft Copilot Studio Exists
Before Copilot Studio, building a conversational AI agent required significant development effort: you needed to write code for natural language understanding (NLU), dialog management, and integration with backend systems. Microsoft Copilot Studio eliminates this complexity by providing a visual authoring canvas, pre-built connectors, and AI-powered features. It exists to democratize bot building—allowing subject matter experts (like sales managers or HR specialists) to create intelligent assistants without needing a team of developers. For the AI-900 exam, you must understand that Copilot Studio is the tool for extending Microsoft Copilot (the out-of-the-box assistant) with custom scenarios and data.
How It Works Internally
Copilot Studio operates on a topic-based architecture. A topic represents a specific conversation flow or a user intent. When a user sends a message, the copilot's AI engine performs the following steps:
Trigger Recognition: The engine analyzes the user's input and matches it to a trigger phrase defined for a topic. For example, if a topic has trigger phrases like "open an account" or "new account," the engine directs the conversation to that topic.
2. Topic Execution: Once triggered, the topic executes a series of nodes (actions) that you have defined in the authoring canvas. Nodes can include: - Message: Sends a text or adaptive card response. - Question: Asks the user for input and stores it in a variable. - Condition: Branches the conversation based on variable values. - Call an action: Invokes a Power Automate flow, a custom connector, or an HTTP request. - Generative Answers: Uses Azure OpenAI to generate a response based on your data sources (like SharePoint, websites, or custom documents).
Generative AI Fallback: If no topic matches the user's input, the copilot can use generative AI to produce an answer. This is enabled by the "Generative Answers" feature, which searches your indexed data and returns a natural language response. You can configure which data sources to use (e.g., SharePoint sites, Microsoft Dataverse, or custom content).
Escalation: If the copilot cannot answer (e.g., confidence is low or user requests a human), it can escalate to a live agent via Omnichannel for Customer Service or a custom integration.
Key Components and Defaults
Topics: Each topic has: - Trigger phrases: At least one phrase is required. You can add multiple trigger phrases (e.g., 5-10) to improve recognition. - Entities: Extract specific pieces of information from user input, like dates, product names, or numbers. Pre-built entities (e.g., Money, DateTime) are available, and you can create custom entities. - Variables: Store user inputs and intermediate values. Variables can be of types: Text, Number, Boolean, Table, etc.
Generative Answers: This feature is powered by Azure OpenAI's GPT models. By default, it uses the Microsoft-managed GPT-3.5 model. You can configure data sources from:
SharePoint sites
OneDrive for Business
Microsoft Dataverse
Custom websites (via web crawling)
Uploaded documents (PDF, Word, Excel, etc.)
Channels: Copilots can be published to:
Microsoft Teams
Custom websites (via a JavaScript snippet or iFrame)
Mobile apps (via direct line API)
Facebook Messenger, Slack, and other third-party channels (via connectors)
Microsoft Dynamics 365 Omnichannel for Customer Service
Authentication: By default, copilots allow anonymous access. You can require authentication using Azure AD, Microsoft Entra ID, or any OAuth2 provider. Authentication is needed to access user-specific data or to enforce security.
Analytics: Copilot Studio provides built-in analytics dashboards showing:
Sessions count
Resolution rate (percentage of conversations resolved without escalation)
Average conversation duration
Topic usage and abandonment rates
Configuration and Verification Commands
Copilot Studio is entirely GUI-based, so there are no CLI commands. However, you can use the following actions to verify and configure:
Test your copilot: Use the "Test Copilot" pane in the authoring canvas to simulate conversations. You can see which topic is triggered and the path taken.
Publish: After testing, click "Publish" to deploy the latest version to all channels. You can also create multiple versions and use "Publish latest version" or "Publish specific version."
Monitor: In the "Analytics" section, you can view real-time and historical data. Use filters by date range, channel, or topic.
Manage data sources: In "Settings" > "Generative AI," you can add or remove data sources for generative answers.
Interaction with Related Technologies
Copilot Studio integrates deeply with the Microsoft Power Platform: - Power Automate: You can call Power Automate flows from within a topic to perform actions like sending an email, creating a record, or processing data. - Power BI: You can embed Power BI reports in copilot responses using adaptive cards. - Microsoft Dataverse: Copilot can read and write data from Dataverse tables, enabling persistent data storage. - Azure OpenAI: The generative AI capabilities rely on Azure OpenAI Service, but you do not need to manage the model yourself. - Microsoft Copilot for Microsoft 365: Copilot Studio allows you to extend the out-of-the-box Microsoft Copilot with custom topics and data sources. This is a key exam concept: Microsoft Copilot is the pre-built assistant, while Copilot Studio is the tool to customize and extend it.
Exam-Important Details
Copilot Studio is a low-code platform, not no-code. Some actions (like calling custom APIs) may require minimal code.
It supports multiple languages for both the authoring interface and the copilot's responses.
The free trial allows you to create copilots with limited capabilities (e.g., up to 300 sessions per month). Full access requires a license (standalone or as part of Microsoft 365 Copilot).
Copilot Studio copilots can be embedded in Microsoft Copilot for Microsoft 365 as custom extensions. This is called "extending Microsoft Copilot."
The Generative Answers feature can be toggled on/off per topic. When enabled, it uses your specified data sources to generate answers.
Security: You can configure data loss prevention (DLP) policies in Power Platform admin center to control which connectors and data sources can be used.
Step-by-Step: Creating a Copilot in Copilot Studio
Create a Copilot: In the Copilot Studio portal, click "Create" and choose a name and language (e.g., English). You can also select a base template (e.g., Customer Support, Employee Onboarding).
Add Topics: Click "Topics" > "New topic" and define trigger phrases. For example, for a "Check Order Status" topic, add phrases like "Where is my order?", "Track order", "Order status".
Design Conversation Flow: Drag nodes onto the canvas: start with a Question node asking for order number, then a Condition node to validate the number, then a Message node to display status. Use Power Automate to fetch data from a backend system.
Enable Generative Answers: Go to "Settings" > "Generative AI" and add data sources (e.g., a SharePoint site with FAQs). Test the copilot with queries outside your topics.
Publish and Deploy: Click "Publish" and then "Deploy" to a channel like Teams or a custom website. Copy the generated code snippet for website embedding.
Monitor and Improve: Use the Analytics dashboard to see which topics are used most, where users drop off, and which queries go unanswered. Update topics and data sources accordingly.
Exam Focus: What AI-900 Tests
The AI-900 exam objective 5.4 specifically tests your understanding of Microsoft Copilot Studio's purpose and capabilities. Key points: - Objective: Describe capabilities of Microsoft Copilot Studio. - Common Wrong Answers:
Confusing Copilot Studio with Microsoft Copilot for Microsoft 365 (the pre-built assistant). Remember: Copilot Studio is the builder tool; Microsoft Copilot is the out-of-the-box assistant.
Thinking Copilot Studio requires coding. It is low-code, not no-code, but main features are drag-and-drop.
Believing Copilot Studio only works for customer service. It can be used for HR, IT, sales, and more.
Specific Terms: "Topics," "Trigger phrases," "Generative Answers," "Power Automate flows," "Channels."
Edge Cases:
Copilot Studio can be used to extend Microsoft Copilot (the one in Teams, Bing, etc.).
You can use custom data sources for generative answers, not just Microsoft 365 data.
Authentication is optional but recommended for sensitive data.
Elimination Strategy: If a question says "built-in copilot that comes with Microsoft 365" — that's Microsoft Copilot. If it says "custom copilot built by users using drag-and-drop" — that's Copilot Studio.
Create a New Copilot
Open Microsoft Copilot Studio and sign in with your work or school account. Click 'Create' on the left navigation. You will be prompted to name your copilot (e.g., 'SupportBot') and select a language (default is English). You can also choose a template: 'Customer Support' pre-populates topics like 'Check Order Status' and 'Return Policy'. For a blank copilot, select 'Create without template'. This step sets the base language and environment. Note: The language you choose affects the AI model's default responses and trigger phrase matching.
Define Topics and Trigger Phrases
Navigate to the 'Topics' tab. Click 'New topic' and give it a name (e.g., 'Greeting'). In the 'Trigger phrases' box, enter phrases a user might say to start this topic, such as 'Hello', 'Hi', 'Good morning'. You can add up to 10 trigger phrases per topic. The AI engine uses these to match user input. For better accuracy, include variations and common misspellings. Then, design the conversation flow by adding nodes from the toolbar: Message, Question, Condition, etc. Each node can be configured with text, variables, and logic.
Integrate Generative AI Responses
Go to 'Settings' > 'Generative AI'. Toggle 'Enable generative answers' on. Under 'Data sources', add sources like SharePoint sites, OneDrive folders, or custom URLs. The copilot will use these to answer questions not covered by topics. You can also set a 'Fallback' behavior: if no topic matches and generative answers fail, you can escalate to a human agent. Test this by asking a question like 'What is your return policy?' — if a topic exists, it triggers; otherwise, generative AI searches your data.
Add Actions with Power Automate
To perform backend operations (e.g., look up order status, create a ticket), you need a Power Automate flow. In the topic canvas, add a 'Call an action' node. Click 'Create a flow' to open Power Automate. Design your flow (e.g., HTTP request to an API, then parse JSON). Save and return to Copilot Studio. The flow appears as an action you can call. Pass variables from the copilot (like order number) to the flow. This enables dynamic data retrieval.
Publish and Deploy to Channels
Once your copilot is complete and tested, click 'Publish' in the top right. Then go to 'Channels'. For Teams deployment, select 'Teams' and follow the prompts to add the copilot as a Teams app. For a website, select 'Custom website' and copy the generated JavaScript snippet. Embed this in your HTML page. You can also deploy to Facebook Messenger, Slack, or Dynamics 365 Omnichannel. After deployment, monitor usage in the 'Analytics' dashboard.
Enterprise Scenario 1: Customer Support for an E-commerce Company
A large online retailer uses Copilot Studio to build a customer support copilot that handles order inquiries, returns, and FAQs. The copilot is deployed on their website and in Microsoft Teams for internal support. Topics include 'Track Order', 'Start a Return', and 'Payment Issues'. The generative AI feature is configured with a SharePoint site containing product manuals and return policies. The copilot handles 80% of inquiries without human intervention. When a customer asks a complex question, the copilot escalates to a live agent via Omnichannel for Customer Service. The company monitors analytics to identify common unresolved issues and updates topics accordingly. A key challenge was ensuring the generative AI responses were accurate; they had to refine the data sources and add custom topics for edge cases.
Enterprise Scenario 2: HR Employee Self-Service
A multinational corporation creates an HR copilot for employees to ask about benefits, leave policies, and payroll. The copilot is deployed in Microsoft Teams. Topics include 'Check PTO Balance', 'Enroll in Health Insurance', and 'Submit Expense Report'. The copilot integrates with Power Automate to query Dataverse for leave balances and trigger approval workflows. Generative AI is enabled with a OneDrive folder containing HR policy PDFs. The copilot is authenticated using Azure AD so that only employees can access it and see their personal data. Performance considerations: the copilot must handle peak usage during open enrollment periods. The company uses load testing and scales by adding more copilot instances. Misconfigurations: initially, the trigger phrases were too narrow, causing many unrecognized queries. They expanded trigger phrases and added synonyms.
Enterprise Scenario 3: IT Help Desk
An IT department builds a copilot to assist with password resets, software requests, and troubleshooting. The copilot is deployed on the company intranet. Topics include 'Reset Password', 'Request Software', and 'Connect to VPN'. The copilot uses Power Automate to execute password resets via Azure AD and to create tickets in ServiceNow. Generative AI is configured with a custom website containing IT knowledge base articles. The copilot reduces IT ticket volume by 30%. A common issue was that the copilot would sometimes provide outdated information from cached data sources; they set up regular refresh schedules for the data sources. They also implemented DLP policies to prevent sensitive data from being shared in generative answers.
Exactly What AI-900 Tests on This Topic (Objective 5.4)
The exam objective is 'Describe capabilities of Microsoft Copilot Studio'. You will be asked to identify what Copilot Studio is used for, its key features, and how it differs from Microsoft Copilot. Specific areas:
Purpose: Low-code platform to create custom copilots.
Key features: Topics, trigger phrases, generative AI, Power Automate integration, multiple channels.
Difference from Microsoft Copilot: Microsoft Copilot is the pre-built AI assistant in Microsoft 365; Copilot Studio is used to extend it or build standalone copilots.
Most Common Wrong Answers and Why
'Copilot Studio is the built-in copilot in Microsoft 365.' Wrong: That's Microsoft Copilot (formerly Bing Chat Enterprise). Copilot Studio is the authoring tool.
'Copilot Studio requires coding in C# or Python.' Wrong: It is low-code; you can build without code, though Power Automate flows may require some logic.
'Copilot Studio only works for customer service scenarios.' Wrong: It is used for any conversational AI scenario (HR, IT, sales, etc.).
'Generative AI in Copilot Studio uses your own Azure OpenAI model.' Wrong: By default, it uses Microsoft-managed GPT-3.5; you can bring your own model but that is not the default.
Specific Numbers, Values, and Terms
Trigger phrases: Up to 10 per topic (though you can add more by editing JSON).
Channels: Teams, custom website, Facebook Messenger, Slack, Dynamics 365 Omnichannel.
Default model: GPT-3.5 (Microsoft managed).
Free trial limit: 300 sessions per month.
Data sources: SharePoint, OneDrive, Dataverse, custom websites, uploaded documents.
Edge Cases and Exceptions
Copilot Studio can extend Microsoft Copilot (the one in Teams, Bing, etc.) by adding custom topics and data sources. This is a common exam scenario.
Authentication is not required but is recommended for personalized experiences.
You can use Copilot Studio to create a copilot that works offline? No, it requires internet connectivity.
Copilot Studio supports multiple languages, but the authoring interface is available in limited languages.
How to Eliminate Wrong Answers
If a question mentions 'pre-built assistant that comes with Microsoft 365', that is Microsoft Copilot. If it mentions 'customizable, drag-and-drop interface', that is Copilot Studio. If it says 'requires no coding', that is misleading—Copilot Studio is low-code, not no-code. Look for keywords: 'topics', 'trigger phrases', 'Power Automate', 'channels'. For generative AI questions, remember that Copilot Studio uses Azure OpenAI but you don't manage the model unless you bring your own.
Copilot Studio is a low-code platform for building custom conversational AI agents (copilots).
Topics and trigger phrases define conversation flows; up to 10 trigger phrases per topic.
Generative Answers feature uses Azure OpenAI (GPT-3.5 by default) to answer questions from your data sources.
Copilot Studio integrates with Power Automate, Dataverse, and multiple channels (Teams, web, etc.).
It can extend Microsoft Copilot for Microsoft 365 with custom topics and data sources.
Authentication is optional; use Azure AD for personalized experiences.
Free trial limited to 300 sessions per month; full license required for production.
Key exam distinction: Microsoft Copilot = pre-built assistant; Copilot Studio = builder tool.
These come up on the exam all the time. Here's how to tell them apart.
Microsoft Copilot (for Microsoft 365)
Pre-built AI assistant integrated into Microsoft 365 apps (Word, Excel, Teams, etc.).
Cannot be customized or extended by end users (only by Microsoft).
Uses Microsoft Graph and enterprise data by default.
No drag-and-drop topic creation; it uses a fixed set of capabilities.
Licensed separately or as part of Microsoft 365 Copilot subscription.
Microsoft Copilot Studio
Low-code platform to build custom copilots from scratch or extend Microsoft Copilot.
Fully customizable: you define topics, trigger phrases, and conversation flows.
Can be connected to any data source via Power Automate or custom connectors.
Visual authoring canvas with drag-and-drop nodes.
Licensed as a standalone product or as part of Microsoft 365 Copilot (for extending).
Mistake
Microsoft Copilot Studio is the same as Microsoft Copilot for Microsoft 365.
Correct
They are different. Microsoft Copilot for Microsoft 365 is the pre-built AI assistant integrated into Microsoft 365 apps. Copilot Studio is a low-code platform to build custom copilots or extend Microsoft Copilot.
Mistake
You must know how to code to use Copilot Studio.
Correct
Copilot Studio is a low-code platform. Most features are drag-and-drop. You can create functional copilots without writing code, though integrating with Power Automate or custom APIs may require minimal coding.
Mistake
Generative AI in Copilot Studio can only use data from Microsoft 365.
Correct
You can configure generative AI to use data from SharePoint, OneDrive, custom websites, Dataverse, and uploaded documents. It is not limited to Microsoft 365 data.
Mistake
Copilot Studio copilots can only be deployed to Microsoft Teams.
Correct
Copilot Studio supports multiple channels: Teams, custom websites, Facebook Messenger, Slack, Dynamics 365 Omnichannel, and more via connectors.
Mistake
Copilot Studio is only for customer support scenarios.
Correct
While customer support is a common use case, Copilot Studio can be used for any conversational AI need: HR, IT, sales, education, etc.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Microsoft Copilot (for Microsoft 365) is the pre-built AI assistant that helps you in Word, Excel, Teams, etc. It is ready to use and cannot be customized. Microsoft Copilot Studio is a low-code platform that lets you build your own custom copilots or extend Microsoft Copilot with custom topics and data sources. Think of Microsoft Copilot as the finished product, and Copilot Studio as the tool to create or modify that product.
No, Copilot Studio is designed as a low-code platform. You can create copilots using drag-and-drop nodes and visual interfaces. However, to integrate with external systems or perform complex logic, you may need to use Power Automate, which can involve some configuration or minimal code. The exam focuses on its low-code nature, so remember that coding is not required for basic copilot creation.
Yes. You can configure generative AI to use data from SharePoint sites, OneDrive folders, Microsoft Dataverse, custom websites, or uploaded documents. This allows the copilot to answer questions based on your specific content. The default AI model is Microsoft-managed GPT-3.5, but you can also bring your own Azure OpenAI model if needed.
You can deploy to Microsoft Teams, custom websites (via JavaScript snippet or iFrame), Facebook Messenger, Slack, and Microsoft Dynamics 365 Omnichannel for Customer Service. You can also use the Direct Line API to build custom channels. The exam may ask which channels are supported; remember Teams and custom websites as the most common.
Copilot Studio is available as a standalone subscription or as part of Microsoft 365 Copilot, which includes the ability to extend Microsoft Copilot. There is also a free trial with limited sessions (300 per month). For full production use, a paid license is required. The exam does not require deep licensing knowledge, but be aware that it is not free.
Absolutely. Copilot Studio is not limited to customer service. You can build copilots for HR, IT, sales, or any other department. For example, an HR copilot can answer employee questions about benefits, leave policies, and payroll, and can integrate with Power Automate to perform actions like checking PTO balances.
Generative Answers is a feature that allows the copilot to answer questions not covered by predefined topics. It uses Azure OpenAI's GPT models to generate responses based on data sources you configure (e.g., SharePoint, websites). This helps handle unexpected queries and reduces the need to create topics for every possible question.
You've just covered Microsoft Copilot Studio — now see how well it sticks with free AI-900 practice questions. Full explanations included, no account needed.
Done with this chapter?