Building AI Agents
with Microsoft Copilot

Follow every step exactly. This builds a real, working AI agent.

STEP 1: ACCESS COPILOT STUDIO
1. Open your browser — go to https://copilotstudio.microsoft.com
2. Sign in with your Microsoft 365 work account
3. If first time: Accept the terms of service and complete setup wizard
4. You land on the Copilot Studio home page — you will see 'My Agents' panel on the left
5. Confirm your environment (top right dropdown) — make sure you are in your correct Microsoft environment (e.g., 'Default' or your organization's environment)

STEP 2: CREATE A NEW AGENT
1. Click the '+ New agent' button (top left or center of the page)
2. Two creation options appear:
Option A: 'Build it yourself' (recommended — full control)
Option B: 'Describe your agent' (AI generates a draft based on your description — good starting point)
3. Click 'Build it yourself'
4. Give your agent a NAME — be specific: 'HR Policy Assistant' not just 'My Bot'
5. Add an optional ICON — helps users identify it in Teams
6. Click 'Create'
7. You are now inside the Copilot Studio agent builder — 5 main sections: Overview, Topics, Knowledge, Actions, Publish

STEP 3: CONFIGURE AGENT PERSONA & INSTRUCTIONS
1. Click 'Overview' in the left sidebar
2. Under 'Instructions' — this is your agent's system prompt. This is CRITICAL.
3. Click into the Instructions field and write your agent's core behavior:

Example Instructions for an HR Policy Agent:
'You are the HR Policy Assistant for [Company Name]. Your role is to answer employee questions about company policies, benefits, time off, and workplace guidelines. Always be professional, empathetic, and accurate. Only answer questions you can find in the provided knowledge sources. If a question is outside your scope or requires personal HR consultation, direct the user to contact HR@company.com or call extension 4400. Never speculate about policies not in the knowledge base. Always cite which policy document your answer comes from.'

4. Under 'Conversation style' — choose:
- More creative (for customer-facing, friendly agents)
- More balanced (for internal business agents — recommended)
- More precise (for compliance and legal agents)
5. Under 'Behavior' — toggle on/off:
- Enable generative answers (AI answers from knowledge — ENABLE this)
- Content moderation level (choose 'Moderate' for most use cases)
6. Click 'Save' after each change

STEP 4: ADD KNOWLEDGE SOURCES
Knowledge sources are what your agent 'knows.' The more relevant content you add, the better it answers.

1. Click 'Knowledge' in the left sidebar
2. Click '+ Add knowledge'
3. Choose source type:

Option A — Public Website:
- Enter the URL of a public website
- Copilot Studio crawls it automatically
- Good for: Public FAQs, product pages, public policies
- Example: https://yourcompany.com/hr/policies

Option B — SharePoint Site:
- Click 'SharePoint' → Enter your SharePoint URL
- Copilot Studio indexes all documents in that library
- Best for: Internal policy docs, handbooks, SOPs
- Requirement: Agent must have access to the SharePoint site

Option C — Upload Files (PDF, Word, Excel, PowerPoint)
- Click 'Files' → Upload documents directly
- Maximum file size: 512 MB per file
- Supported formats: PDF, DOCX, XLSX, PPTX, TXT
- Best for: Specific documents you want the agent to reference

Option D — Dataverse Tables
- Connect to your organization's Dataverse database
- Agent can query structured data in real-time
- Best for: Product catalogs, employee directories, case databases

4. After adding each source, click 'Refresh' to let Copilot Studio index the content
5. Test that knowledge was indexed: Type a question in the Test pane that should be answered by your document

STEP 5: CREATE CUSTOM TOPICS
Topics are guided conversation flows for specific, structured interactions.

When to use Topics vs. Generative Answers:
- Generative answers: Open-ended Q&A from documents (most questions)
- Topics: Structured flows where you need to collect data, make decisions, or trigger actions

Example topic to create: 'Submit Time Off Request'
1. Click 'Topics' in left sidebar
2. Click '+ Add topic' → 'From blank'
3. Name it: 'Submit Time Off Request'
4. Add trigger phrases (what users say to start this topic):
- 'I want to request time off'
- 'How do I submit a vacation request'
- 'PTO request'
- 'I need a day off'
- 'Request leave'
5. Click the canvas to start building the flow:
Node 1 — Message: 'I can help you submit a time off request. Let me collect a few details.'
Node 2 — Question: 'What type of leave do you need?' (options: Vacation, Sick, Personal, Bereavement)
Node 3 — Question: 'What is your start date?' (save to variable: StartDate)
Node 4 — Question: 'What is your return date?' (save to variable: EndDate)
Node 5 — Question: 'Would you like to add a note for your manager?' (save to variable: Note)
Node 6 — Action: Trigger Power Automate flow 'Submit Leave Request' (pass: StartDate, EndDate, Note)
Node 7 — Message: 'Your time off request for [StartDate] through [EndDate] has been submitted. Your manager will receive a notification.'
6. Click 'Save topic'
7. Test in the Test pane by typing 'I want time off'

STEP 6: ADD ACTIONS (Connect to Real-World Systems)
Actions transform your agent from a chatbot into an AI agent that gets things done.

1. Click 'Actions' in the left sidebar
2. Click '+ Add action'
3. Three action types:

Type A — Power Automate Flow (Recommended for most scenarios):
1. Click 'Power Automate'
2. Click 'Create a flow' — this opens Power Automate in a new tab
3. The flow automatically starts with 'Run a flow from Copilot Studio' trigger
4. Add your automation steps:
Example — Send Email action:
- Add action: 'Send an email (V2)' from Office 365 Outlook connector
- To: Use the variable passed from the agent (e.g., manager email)
- Subject: 'Time Off Request from [employee name]'
- Body: Compose with the variables from the agent conversation
5. Add 'Return values to Copilot Studio' as the last step (outputs what the agent should know happened)
6. Save and publish the flow
7. Return to Copilot Studio — your flow now appears as an available action
8. Add input parameters (what agent sends to the flow) and output parameters (what flow returns to agent)

Type B — HTTP Request (Call any REST API):
1. Click 'HTTP'
2. Enter API endpoint URL
3. Select method: GET / POST / PUT / DELETE
4. Add headers (Authorization, Content-Type)
5. Add request body with agent variables mapped to API parameters
6. Map API response fields to agent variables
Example: Call a weather API, a CRM API, or a ticketing system API

Type C — Connector Action (Pre-built integrations):
1. Click 'Connector'
2. Browse 1,000+ pre-built connectors: ServiceNow, Salesforce, SAP, Zendesk, Jira, etc.
3. Select your connector → authenticate → select the action
4. Map agent variables to connector parameters
Example: Create ServiceNow ticket, update Salesforce record, post to Slack

STEP 7: SET UP AUTHENTICATION
Authentication determines who can use your agent and what data it can access.

1. Click 'Settings' (gear icon, top right of agent builder)
2. Click 'Security' tab → 'Authentication'
3. Choose authentication type:

Option A — No authentication:
- Anyone with the link can use the agent
- Good for: Public-facing customer support
- Limitation: Agent cannot access user-specific data

Option B — Authenticate with Microsoft (Recommended for internal agents):
- Users sign in with their Microsoft 365 account
- Agent can access user-specific data (name, email, department, calendar)
- Uses Microsoft Entra ID (formerly Azure AD)
- Steps:
1. Select 'Authenticate with Microsoft'
2. Click 'Edit' to configure the app registration
3. In Azure Portal: Create an App Registration for your agent
4. Copy Application (client) ID and paste into Copilot Studio
5. Add API permissions: Microsoft Graph (User.Read, Mail.Read as needed)
6. Create a client secret and paste into Copilot Studio
7. Click 'Save' and test authentication

Option C — Manual (OAuth 2.0 for external systems):
- For agents that connect to non-Microsoft external services
- Configure OAuth 2.0 settings manually

STEP 8: CONFIGURE GENERATIVE AI SETTINGS
1. Click 'Settings' → 'Generative AI'
2. Content moderation: Set to 'Moderate' (blocks harmful content, allows normal business use)
3. Knowledge grounding: Toggle ON 'Only answer from knowledge sources' — this prevents hallucination
4. Response format: Choose whether agent can use bullet points, bold text, tables in responses
5. Sensitive information: Toggle ON 'Avoid displaying sensitive data'
6. Citation style: Choose whether agent cites source documents in answers (recommended: ON)
7. Click 'Save'

STEP 9: TEST YOUR AGENT THOROUGHLY
1. Click 'Test' button (top right) — opens the Test pane on the right
2. Run through this testing checklist:

TEST 1 — Knowledge answers:
Type a question from your knowledge source documents
Expected: Accurate answer citing the source document
If failing: Check that knowledge source is indexed (refresh it)

TEST 2 — Topic triggers:
Type each trigger phrase for your custom topics
Expected: Topic flow starts immediately
If failing: Add more trigger phrase variations

TEST 3 — Action execution:
Trigger a topic that calls a Power Automate flow
Expected: Flow runs, confirmation message returned
If failing: Check flow in Power Automate for errors

TEST 4 — Fallback behavior:
Ask something completely unrelated to your agent's scope
Expected: 'I'm sorry, I can only help with [scope]. Please contact [support] for other questions.'
If failing: Edit your fallback topic with a better response

TEST 5 — Edge cases:
Try: Extremely long messages, questions in another language, gibberish input
Expected: Graceful handling in all cases

3. Click 'Reset' between tests to start fresh conversations
4. Click the bug icon to see detailed execution logs for debugging

STEP 10: PUBLISH YOUR AGENT
Publishing makes your agent available to users through one or more channels.

1. Click 'Publish' in the left sidebar
2. Click 'Publish' button — this publishes the latest version of your agent
3. Status changes to 'Published' — note the version number for your records
4. Choose your channels to publish to:

CHANNEL A — Microsoft Teams:
1. Under 'Channels' → Click 'Microsoft Teams'
2. Click 'Turn on Teams' to enable
3. Configure:
- App Name: What it's called in Teams
- Short Description: What users see when they find it
- Long Description: Full description for Teams app catalog
- App icon: Upload your branded icon (192x192 px recommended)
4. Click 'Submit for admin approval' (or if you're admin: Add to Teams immediately)
5. IT Admin approves in Teams Admin Center → Microsoft 365 Apps
6. Agent appears as a personal app in Teams for all users (or targeted group)

CHANNEL B — SharePoint Website:
1. Under 'Channels' → Click 'SharePoint'
2. Click 'Add to SharePoint'
3. Choose the SharePoint site
4. Copy the agent web part
5. In SharePoint: Edit your page → Add web part → Paste the Copilot Studio web part
6. Configure position and appearance
7. Publish the SharePoint page

CHANNEL C — Public Website:
1. Under 'Channels' → Click 'Demo website' or 'Custom website'
2. Copy the JavaScript embed code snippet
3. Paste into your website's HTML before the closing body tag
4. The agent chat widget appears in the bottom-right corner of your website

CHANNEL D — Microsoft 365 Copilot (BizChat):
1. Under 'Channels' → Click 'Microsoft 365'
2. Toggle 'Enable agent in Microsoft 365 Copilot'
3. IT Admin must approve in Microsoft 365 Admin Center
4. Agent appears as a plugin in BizChat — users can invoke it by @mentioning it

STEP 11: CONFIGURE ESCALATION TO HUMANS
Even the best agents need human backup.

1. Click 'Topics' → find 'Escalate' (pre-built system topic)
2. Edit this topic:
- Add a message: 'Let me connect you with a specialist who can help.'
- Add 'Transfer conversation' action
- Configure handoff to: Teams, Dynamics 365 Omnichannel, or custom webhook
3. Create escalation triggers in your custom topics:
- Add a Condition: If user expresses frustration → escalate
- Keywords: 'speak to a person', 'human', 'representative', 'agent', 'frustrated'
4. Set up after-hours handling:
- Add a condition checking current time
- If outside business hours: 'Our agents are available Mon–Fri 9am–5pm EST. I have logged your request and someone will follow up by next business day.'

STEP 12: MONITOR PERFORMANCE & IMPROVE
Publishing is not the end — it is the beginning.

1. Click 'Analytics' in the left sidebar
2. Review these key metrics weekly:

ENGAGEMENT METRICS:
- Total sessions: Is the agent being used?
- Sessions over time: Growing or declining?
- Unique users: How many different people?

PERFORMANCE METRICS:
- Resolution rate: Target 70%+ without human handoff
- Escalation rate: Should be under 30%
- Abandonment rate: Users who left mid-conversation

QUALITY METRICS:
- Customer satisfaction (CSAT): If survey enabled
- Top topics: Most common requests
- Unrecognized inputs: Questions the agent couldn't categorize

3. Click 'Transcripts' to read actual user conversations
- Look for: Incorrect answers, confused users, missing knowledge
- Take notes on: What topics users ask that you haven't built yet

4. Improve your agent monthly:
- Add new knowledge sources for unanswered questions
- Create new topics for common unhandled requests
- Refine trigger phrases based on how users actually phrase questions
- Update system instructions based on conversation quality

5. Set up automated alerts (via Power Automate):
- Alert when resolution rate drops below 70%
- Alert when new negative CSAT scores are received
- Weekly performance summary emailed to agent owner

SharePoint Agents are the fastest way to make your organizational documents conversational. Users ask questions in plain English and the agent answers from your files.

WHAT YOU NEED:
- SharePoint Premium license (or Microsoft 365 Copilot)
- A SharePoint document library with your content
- Site Owner or Site Admin permissions

STEP-BY-STEP BUILD:

Step 1: Prepare Your SharePoint Library
1. Open your SharePoint site
2. Go to the document library you want to use as the knowledge base
3. Ensure documents are well-organized with clear, descriptive filenames
4. Make sure all relevant documents are in the library (PDFs, Word docs, Excel files all supported)
5. Check that documents have been recently modified or uploaded (SharePoint indexes recent content first)
6. Recommended: Create a dedicated 'AI Knowledge Base' folder or library for cleaner scope

Step 2: Create the Agent
1. In your SharePoint document library, click the 'Copilot' icon (sparkle icon in top toolbar)
OR: Click the Settings gear → 'Add an agent'
OR: Click '+ New' → 'Agent'
2. The agent creation pane opens on the right
3. Give it a name: 'HR Policies Agent' or 'Employee Handbook Agent'
4. Under 'Sources': Your current library is automatically added as the knowledge source
5. Optional: Add additional SharePoint sites or libraries as additional sources
6. Click 'Create'

Step 3: Configure the Agent
1. Click 'Edit' to open the agent configuration
2. Add a description: What is this agent for?
3. Write starter questions (what users might ask):
- 'How many days of PTO do I get?'
- 'What is the remote work policy?'
- 'How do I submit an expense report?'
4. These appear as clickable chips in the agent interface — reduce friction for new users
5. Click 'Save'

Step 4: Add the Agent to Your SharePoint Page
1. Go to the SharePoint page where you want the agent
2. Click 'Edit' (top right)
3. Click '+ Add a new web part' in the page canvas
4. Search for 'Copilot' or 'Agent'
5. Select your agent from the list
6. Position the web part where you want it on the page
7. Click 'Publish' to make the page live

Step 5: Share the Agent
Option A — SharePoint page: Share the page URL with your organization
Option B — Teams: Share the agent URL → opens in Teams web tab
Option C — Email: Copy agent link and email to your team
Option D — Teams Channel Tab: Add as a tab in your Teams channel

Step 6: Test Your SharePoint Agent
1. Ask 5 questions you know the answers to from your documents
2. Verify the agent cites the correct source documents
3. Ask something NOT in your documents — agent should say it cannot find that information
4. Ask follow-up questions in the same conversation — agent should maintain context

KEY LIMITATIONS TO KNOW:
- Only works with content in SharePoint (cannot browse the internet)
- Requires SharePoint Premium or M365 Copilot for full functionality
- Agent answers are only as good as your documents — keep them updated
- Does not currently support actions/integrations (knowledge Q&A only)

Teams Agents are AI agents built directly into a Teams team or channel. They can answer questions, trigger actions, and serve as the team's AI-powered assistant.

THREE WAYS TO ADD AN AGENT TO TEAMS:

METHOD 1: Deploy a Copilot Studio Agent to Teams (Full Featured)
This is the most capable option — covered fully in Module 2, Step 10.

METHOD 2: Use Microsoft 365 Copilot BizChat in Teams
1. Open Teams
2. Click 'Copilot' icon in the left sidebar (the sparkle icon)
3. This is BizChat — your personal AI agent in Teams
4. It can access all your M365 data: emails, files, calendar, meetings
5. Ask it anything: 'What were the action items from my last meeting?'
6. Use it to: Draft messages, summarize channels, find files, prepare for meetings

METHOD 3: Create a Team-Specific Agent (NEW in 2025)
1. Open the Teams team where you want to add an agent
2. Click '+' next to the channel tabs
3. Search for 'Copilot Studio' or 'Agent'
4. Select your published Copilot Studio agent
5. The agent now appears as a tab in that channel
6. Team members can interact with the agent directly in the channel

STEP-BY-STEP: CONFIGURE TEAMS AGENT FROM COPILOT STUDIO

Step 1: Build your agent in Copilot Studio (follow Module 2 steps)

Step 2: Configure for Teams
1. In Copilot Studio → Click 'Channels' → 'Microsoft Teams'
2. Click 'Edit details' to configure the Teams app package:
- Short name: What appears in Teams (max 30 characters)
- Full name: Complete name (max 100 characters)
- Short description: One sentence summary (max 80 characters)
- Full description: Full explanation (max 4,000 characters)
- Privacy policy URL: Required for publishing to org
- Terms of use URL: Required for publishing to org
- App icons: Color icon (192x192 px) and Outline icon (32x32 px)
- Accent color: Hex color for the app tile
3. Click 'Save'

Step 3: Publish to Teams
1. Click 'Publish to Teams'
2. If you are a Teams Admin: The agent deploys immediately
3. If not Admin: A request is submitted to your Teams Admin for approval
4. Admin approves in: Teams Admin Center → Manage Apps → Submitted apps

Step 4: Control who can use it
1. Teams Admin Center → Manage Apps → Find your agent → App policies
2. Set to: Everyone, specific groups, or specific users
3. Optional: Pin the agent to Teams left rail for all users (ensures high visibility)

Step 5: Set up proactive messages (Agent reaches out to users)
1. In Copilot Studio → Actions → Power Automate
2. Build a flow that triggers on schedule or event
3. Use the 'Post message in Teams chat (V3)' action
4. Agent sends a message to a user or channel proactively
Example: Every Monday at 8am, agent sends team a summary of open tasks from Planner

TEAMS AGENT BEST PRACTICES:
- Pin high-use agents to the left rail for easy access
- Create a 'getting started' message that fires on first use
- Include a help command: If user types 'help' → show list of what agent can do
- Use buttons and quick replies to reduce typing — mobile users especially appreciate this
- Set up feedback collection: Ask 'Was this helpful?' after every answer

Power Automate transforms your Copilot agent from a chatbot into an AI agent that actually DOES things. This is where the real automation power lives.

HOW THE INTEGRATION WORKS:
Copilot Studio Agent → User requests an action → Agent calls Power Automate flow → Flow executes actions (email, database update, ticket creation, etc.) → Flow returns result to agent → Agent reports result to user

STEP-BY-STEP: BUILD YOUR FIRST AGENT ACTION FLOW

Example: 'Create Support Ticket' action
When a user tells the agent they have a technical issue, the agent creates a ServiceNow ticket automatically.

Step 1: Start in Copilot Studio
1. Open your agent in Copilot Studio
2. Click 'Actions' → '+ Add action'
3. Select 'Power Automate'
4. Click 'Create a new flow'
5. Power Automate opens in a new browser tab

Step 2: Build the Power Automate Flow
1. The flow automatically starts with trigger: 'Run a flow from Copilot'
2. You see two boxes: Inputs (from agent) and Outputs (back to agent)
3. Add inputs — what the agent will send to this flow:
- Input 1: 'IssueDescription' (Text) — what the user described
- Input 2: 'IssuePriority' (Text) — High/Medium/Low
- Input 3: 'UserEmail' (Text) — user's email from authentication
4. Click '+ New step' to add your automation:

CONNECTOR: ServiceNow
Action: 'Create Record'
Table: 'Incidents'
Map fields:
- Short description → IssueDescription input
- Priority → IssuePriority input
- Caller → UserEmail input
- Category → 'Technical'

5. After the ServiceNow action, add another step:
CONNECTOR: Office 365 Outlook
Action: 'Send an email (V2)'
To: UserEmail input
Subject: 'Your support ticket has been created'
Body: 'Your ticket [Ticket Number from ServiceNow output] has been created and assigned to the IT team. Expected response within 4 business hours.'

6. Add 'Respond to Copilot' final step:
Output 1: 'TicketNumber' → Map from ServiceNow create record output
Output 2: 'ConfirmationMessage' → 'Your ticket has been created successfully'

7. Click 'Save'

Step 3: Connect Flow to Your Agent
1. Return to Copilot Studio (Power Automate tab should still be open)
2. Your new flow appears in the Actions list
3. Click on it to configure:
- Set input mappings: IssueDescription → agent variable, IssuePriority → agent variable
- Set output mappings: TicketNumber → create new agent variable
4. Click 'Save'

Step 4: Add the Action to a Topic
1. Go to 'Topics' → Create or edit the 'Report IT Issue' topic
2. Add conversation nodes:
- Message: 'I can create a support ticket for you. Let me get some details.'
- Question: 'Please describe the issue you are experiencing.' (save to variable: IssueDescription)
- Question: 'What is the priority? High, Medium, or Low?' (save to variable: IssuePriority)
- Action: Call your 'Create Support Ticket' Power Automate flow
Pass: IssueDescription, IssuePriority
Store output: TicketNumber
- Message: 'Your ticket has been created! Ticket number: {TicketNumber}. The IT team has been notified and you will receive email confirmation.'
3. Save the topic

Step 5: Test End-to-End
1. In Test pane: Say 'I have a technical problem'
2. Agent asks for description → you provide it
3. Agent asks for priority → you provide it
4. Agent calls the Power Automate flow
5. Flow creates ticket in ServiceNow
6. Flow sends confirmation email
7. Agent reports ticket number to user

TROUBLESHOOTING COMMON ERRORS:
- 'Flow not triggering': Check that the flow is turned ON in Power Automate
- 'Connection error': Re-authorize the connector in Power Automate
- 'Variable not passing': Verify exact variable names match between Copilot Studio and the flow
- 'Timeout error': Add error handling in the flow and return a user-friendly error message

These are the most valuable automation flows to build for your Copilot agents. Each is fully detailed.

FLOW 1: EMAIL TRIAGE & RESPONSE AGENT
Trigger: New email arrives in shared mailbox
Steps:
1. Get email content (Subject, Body, From, Attachments)
2. Send to Azure OpenAI (or Claude via HTTP connector)
Prompt: 'Classify this email as: Client Inquiry / Invoice / Newsletter / Urgent Action / Spam. Also draft a professional 3-sentence response if it is a Client Inquiry. Email: [body]'
3. Parse AI response (extract category and draft)
4. If category = Client Inquiry:
- Create draft email reply with AI-generated response
- Add email to a 'Needs Review' SharePoint list
- Send Teams notification to account manager
5. If category = Invoice:
- Forward to accounting@company.com
- Add to invoice tracking spreadsheet
6. If category = Urgent Action:
- Flag email in Outlook
- Send Teams urgent alert to manager
7. Log all emails to Dataverse for reporting

FLOW 2: MEETING PREP BRIEFING AGENT
Trigger: 30 minutes before any calendar event
Steps:
1. Get meeting details (attendees, subject, description)
2. For each attendee: Search for recent emails from them (last 7 days)
3. Search for recent documents related to meeting topic
4. Send all context to Azure OpenAI
Prompt: 'Create a 1-page meeting prep briefing. Include: meeting purpose, attendee background based on recent communications, key topics likely to come up, and 3 questions I should be prepared to answer.'
5. Send the briefing to yourself via email or Teams message

FLOW 3: DAILY BUSINESS INTELLIGENCE AGENT
Trigger: Scheduled — weekdays at 5:30 PM
Steps:
1. Get today's sales data from CRM (Salesforce, Dynamics 365, or HubSpot)
2. Get marketing metrics (Google Analytics, Meta Ads via HTTP connector)
3. Get email list stats (Mailchimp, ConvertKit via HTTP connector)
4. Get support ticket status from ServiceNow
5. Combine all data and send to Azure OpenAI
Prompt: 'Create a daily business intelligence summary. Include: top wins, areas of concern, KPIs vs. target, and 2 recommended actions for tomorrow.'
6. Send formatted report via email and Teams message

FLOW 4: DOCUMENT GENERATION AGENT
Trigger: Copilot Studio agent request (from user conversation)
Inputs: ClientName, ProjectType, BudgetRange, Timeline
Steps:
1. Get document template from SharePoint based on ProjectType
2. Send to Azure OpenAI with client details
Prompt: 'Generate a professional project proposal for [ClientName]. Project: [ProjectType]. Budget: [BudgetRange]. Timeline: [Timeline]. Use this template as the structure: [template content]'
3. Create new Word document using 'Create file' connector
4. Save to SharePoint in client's folder
5. Share document link with agent to return to user
6. Send notification email to sales team

FLOW 5: APPROVAL WORKFLOW AGENT
Trigger: Copilot Studio agent request or form submission
Steps:
1. Collect request details (purchase request, policy exception, etc.)
2. Create item in SharePoint Approvals list
3. Send approval request to manager via Teams Adaptive Card
Card includes: Request details, Approve/Reject buttons, optional comment field
4. Manager clicks Approve or Reject in Teams
5. If Approved:
- Notify requestor via Teams message
- Update approval log in SharePoint
- Trigger downstream action (create PO, activate exception, etc.)
6. If Rejected:
- Notify requestor with rejection reason
- Log in approval history
7. Agent returns final status to user

FLOW 6: CUSTOMER ONBOARDING AGENT
Trigger: New customer record created in CRM
Steps:
1. Get new customer details from CRM
2. Create folders in SharePoint for new customer
3. Generate welcome email using Azure OpenAI with customer details
4. Send welcome email from shared mailbox
5. Create Teams channel for customer success team
6. Post customer overview in new channel
7. Create 30-day onboarding task checklist in Planner
8. Assign tasks to team members based on role
9. Schedule kickoff meeting in Outlook (check all calendars for availability)
10. Update CRM status to 'Onboarding'

FLOW 7: COMPLIANCE MONITORING AGENT
Trigger: Scheduled weekly
Steps:
1. Scan SharePoint libraries for documents with specific metadata (Compliance Review Date)
2. Filter documents where review date is within 30 days or past due
3. For each document: Generate compliance status summary using Azure OpenAI
4. Create compliance report in Excel
5. Post report to Teams compliance channel
6. Send individual notifications to document owners
7. Escalate overdue items to Compliance Manager

FLOW 8: SOCIAL MEDIA & CONTENT PUBLISHING AGENT
Trigger: Scheduled weekly (Mondays at 7 AM)
Steps:
1. Read content calendar from SharePoint list
2. For each planned post: Send topic to Azure OpenAI
Prompt: 'Write a [LinkedIn/Twitter/Instagram] post about [topic] for [brand name]. Tone: [professional/casual/inspirational]. Include 5 relevant hashtags. Max [character count] characters.'
3. Save generated posts to content calendar SharePoint list
4. Send Teams notification for review
5. If approved (Teams Adaptive Card): Schedule posts via Buffer or Hootsuite HTTP connector
6. If rejected: Flag for manual revision

The system prompt (called 'Instructions' in Copilot Studio) is the single most impactful element of your agent. A well-written system prompt makes an average agent exceptional. A poorly written one makes a capable agent useless.

THE ANATOMY OF A PERFECT COPILOT AGENT SYSTEM PROMPT:

PART 1 — IDENTITY (Who the agent is):
'You are [Agent Name], the [Role] for [Organization]. You were created by [Creator/Department] to serve [Primary User Group].'

PART 2 — PURPOSE (What the agent does):
'Your primary function is to [main goal]. Specifically, you help users with:
- [Task 1]
- [Task 2]
- [Task 3]'

PART 3 — KNOWLEDGE BASIS (What it knows):
'You answer questions based solely on the following knowledge sources: [list sources]. Do not answer from general knowledge. If the answer is not in your knowledge sources, tell the user clearly.'

PART 4 — TONE & STYLE (How it communicates):
'Your communication style is [professional/friendly/formal/empathetic]. You:
- Use clear, concise language
- Avoid jargon unless the user uses it first
- Structure responses with bullet points for lists of 3 or more items
- Begin responses with the most important information
- Never use phrases like: "Certainly!", "Great question!", "I apologize but"'

PART 5 — BOUNDARIES (What it will NOT do):
'You will NOT:
- Answer questions outside of [defined scope]
- Provide [specific type of advice — legal, medical, financial without disclaimer]
- Share information the user does not have permission to access
- Make promises or commitments on behalf of the organization
- Discuss [specific off-limits topics]'

PART 6 — ESCALATION RULES (When to hand off to a human):
'If a user:
- Expresses frustration more than twice in a conversation
- Asks to speak to a person
- Has a question you cannot find in your knowledge sources after 2 attempts
- Mentions an urgent or emergency situation
Then immediately use the escalation action to connect them with [support contact / human team].'

PART 7 — SPECIAL INSTRUCTIONS (Unique behaviors):
'Additional rules:
- Always cite the source document when answering from knowledge
- If asked about policies: Include the policy effective date
- Greet returning users by name if authentication is enabled
- End every conversation by asking: "Is there anything else I can help you with?"'

FULL EXAMPLE — HR Policy Agent System Prompt:
'You are the HR Policy Assistant, the official HR knowledge resource for Acme Corporation. You were created by the HR department to help employees quickly find accurate information about company policies, benefits, and workplace guidelines.

Your primary function is to answer HR-related questions accurately and compassionately. You help employees with: policy questions, benefits information, PTO and leave requests, onboarding questions, performance management processes, and workplace conduct guidelines.

You answer only from the official HR document library in SharePoint. All answers must be grounded in these documents. If you cannot find an answer in the knowledge base, say clearly: "I don't have that information in my current knowledge base. Please contact HR at hr@acme.com or call ext. 4400 for assistance."

Your communication style is warm, professional, and supportive. Employees may come to you with sensitive questions — always respond with empathy and confidentiality. Use plain English. Avoid HR jargon.

You will NOT provide legal advice, make disciplinary decisions, disclose other employees' personal information, or discuss compensation decisions. For sensitive personal matters, always recommend speaking directly with an HR Business Partner.

If an employee expresses distress, urgency, or asks for a human, immediately trigger the escalation to connect them with a live HR representative. The Employee Assistance Program (EAP) number is 1-800-XXX-XXXX — mention this for wellbeing-related concerns.

Always cite the policy name and section when answering policy questions. End each response by checking: Is there anything else I can help you with today?'

WRITING TIPS:
1. Be specific, not general — vague instructions produce vague behavior
2. Use negative instructions ('Do NOT') explicitly — do not assume the agent will infer limits
3. Include examples of good responses if you have them
4. Keep it under 2,000 words — longer prompts sometimes reduce focus
5. Test every instruction — verify each rule actually changes agent behavior
6. Update the prompt every 30 days based on conversation transcript review

As a user of Copilot agents, the quality of your prompts also determines the quality of results. Here are templates for every scenario.

FOR BUSINESS CHAT (BizChat) — PERSONAL PRODUCTIVITY AGENT:

Template 1: Morning Briefing
'Good morning. Please give me a briefing for today. Include: (1) My meetings today with attendee names and key context from recent emails about each meeting. (2) Any urgent emails received in the last 18 hours. (3) Any tasks due today from my To Do list. Format as a structured daily briefing.'

Template 2: Document Synthesis
'I have attached [document name]. Please summarize the 5 most important points. Then identify: (1) Any action items mentioned. (2) Any decisions that need to be made. (3) Any deadlines referenced. Format as an executive summary.'

Template 3: Research Across Your Data
'Search through my recent emails, Teams chats, and files from the past [2 weeks] related to [Project Name]. Give me a comprehensive status update including: what has been discussed, what decisions were made, what is outstanding, and who is responsible for what.'

Template 4: Communication Draft
'Using what you know from my recent email thread with [Name] about [Topic], draft a [formal/casual] follow-up email that: (1) Thanks them for [specific thing]. (2) Updates them on [status]. (3) Requests [specific action]. (4) Proposes a [call/meeting] for [date range]. Keep it under [150] words.'

Template 5: Meeting Preparation
'I have a meeting in [30 minutes] with [Name/Company] about [Topic]. Based on our recent emails, any shared documents, and their LinkedIn summary if available — prepare me a 1-page briefing that includes: their priorities based on recent communications, key questions they may ask me, points I should make, and any open items from our last interaction.'

FOR COPILOT STUDIO AGENTS — OPTIMIZING USER INTERACTIONS:

Template 6: Structured Request
'I need to [action]. Here are the details:
- [Detail 1]
- [Detail 2]
- [Detail 3]
Please [specific output].'

Template 7: Policy Question
'What is the company policy on [topic]? Specifically, I want to know: (1) What is allowed. (2) What is not allowed. (3) What is the process to request an exception. Please cite the policy document.'

Template 8: Process Initiation
'Please [start a process/create a ticket/submit a request] for the following:
- Type: [type]
- Description: [description]
- Urgency: [high/medium/low]
- My preferred contact: [email/phone]'

FOR COPILOT IN TEAMS — MEETING AGENT:

Template 9: Real-Time Meeting Query
During a meeting, type in the Copilot pane:
'What decision was made about [topic] in this meeting so far?'
'Who has been assigned to [task] in this meeting?'
'Summarize what [person] said about [topic]'
'Are there any unresolved questions or disagreements so far?'

Template 10: Post-Meeting Action
'Generate a complete meeting recap for the [meeting name] meeting that just ended. Include: (1) Attendees. (2) Key discussion points (max 5 bullets). (3) All decisions made. (4) All action items with owner names and due dates. (5) Questions that were not resolved. Format for sharing with a team that was not in the meeting.'

Each blueprint includes: what to build, exact tools to use, step-by-step configuration, and what success looks like.

BLUEPRINT 1: GOVERNMENT — Constituent Services Agent
Platform: Copilot Studio + Power Automate
Deploy to: Public website + Microsoft Teams

What it does:
- Answers citizen questions about permits, services, and procedures
- Accepts service requests and routes to correct department
- Provides real-time status updates on submitted requests
- Handles after-hours inquiries with appropriate messaging

Build steps:
1. Copilot Studio: Create agent named 'Citizen Services Assistant'
2. Knowledge: Add public website pages (all public-facing service pages)
3. Knowledge: Upload service guide PDFs
4. Topics: Create 'Submit Service Request' flow
5. Actions: Power Automate flow → Create item in SharePoint list → Email to correct department
6. Actions: Power Automate flow → Query status from case management system
7. Authentication: None (public-facing)
8. Publish: Public website embed + Teams for internal staff use
9. Analytics: Monitor daily for unanswered questions, add to knowledge

Security note: Do not connect to any systems containing classified or sensitive PII. Public-facing agents should only access publicly available information.

---

BLUEPRINT 2: HEALTHCARE — Patient Information Agent
Platform: Copilot Studio + Power Automate (HIPAA-compliant setup required)
Deploy to: Practice website (authenticated patients only)

What it does:
- Answers FAQs about services, hours, insurance, and procedures
- Allows patients to request appointment availability check
- Provides pre-appointment instructions from templates
- Escalates clinical questions to nursing staff immediately

Build steps:
1. Copilot Studio: Create agent with strict scope: NO clinical advice
2. Instructions: 'You provide administrative information only. Never diagnose, treat, or provide medical advice. For any medical questions, always direct to clinical staff.'
3. Knowledge: Office hours, insurance accepted, location info, billing FAQs
4. Topics: 'Request Appointment Information' → Power Automate → Add to callback request list → Email front desk
5. Escalation: Any mention of symptoms, medications, pain → immediately show clinical contact
6. Authentication: Require Microsoft SSO for any personal health information
Critical: Have legal and compliance review before deployment in healthcare settings.

---

BLUEPRINT 3: EDUCATION — Student Support Agent
Platform: Copilot Studio + Power Automate + Teams
Deploy to: Student portal (SharePoint) + Teams

What it does:
- Answers questions about courses, registration, financial aid, campus resources
- Helps students find the right department or office for their question
- Submits support requests (IT help, disability accommodations, advising requests)
- Provides 24/7 coverage when offices are closed

Build steps:
1. Knowledge: Student handbook, course catalog, financial aid guide, campus services directory
2. Topics: 'Submit IT Help Request' → ServiceNow ticket
3. Topics: 'Request Advising Appointment' → Booking system integration
4. Topics: 'Financial Aid Question' → Knowledge Q&A + escalation to financial aid office
5. Authentication: Azure AD (student accounts)
6. Publish: SharePoint student portal + Teams (student team)
7. Add CSAT survey: 'Was this helpful? Yes/No' after every conversation

---

BLUEPRINT 4: RETAIL/E-COMMERCE — Customer Service Agent
Platform: Copilot Studio + Power Automate
Deploy to: Website + Teams (for internal order management)

What it does:
- Handles order status inquiries automatically
- Processes return requests and generates return labels
- Answers product questions from product catalog
- Routes complex complaints to human agents with full context

Build steps:
1. Knowledge: Product FAQ, return policy, shipping policy
2. Topics: 'Check Order Status' → Power Automate → Query Shopify/WooCommerce API → Return status
3. Topics: 'Start a Return' → Collect order number → Validate → Generate return label → Email customer
4. Topics: 'Product Question' → Knowledge base Q&A from product catalog
5. Escalation: Complaints, damaged items, fraud → Route to human with conversation context
6. Authentication: None for basic Q&A; Order number + email for order-specific queries

---

BLUEPRINT 5: PROFESSIONAL SERVICES (Consulting/Law/Accounting) — Client Intake Agent
Platform: Copilot Studio + Power Automate
Deploy to: Firm website + Teams

What it does:
- Conducts structured intake for new client inquiries
- Qualifies prospects based on defined criteria
- Schedules discovery consultations with qualified prospects
- Sends intake summary to relevant attorney/consultant for review

Build steps:
1. Topics: 'New Client Inquiry' — structured intake form:
- What type of help are you looking for? (service area)
- Briefly describe your situation
- How soon do you need assistance?
- What is your approximate budget/timeline?
2. Action: Classify inquiry → Route to correct practice area (Power Automate + OpenAI)
3. Action: If qualified → Book discovery call via Calendly/Bookings API
4. Action: Send intake summary to attorney via Teams message
5. If not qualified: Provide helpful resources and referral information
6. Data: Log all inquiries to Dataverse CRM

---

BLUEPRINT 6: MINISTRY/NON-PROFIT — Community Care Agent (Daily Lift Bible)
Platform: Copilot Studio + Power Automate
Deploy to: Website + potentially SMS via Twilio

What it does:
- Answers questions about events, programs, service times
- Receives prayer requests and routes to prayer team
- Provides devotional content and scripture references
- Connects community members with appropriate ministry resources

Build steps:
1. Knowledge: Church website content, event calendar, ministry descriptions, FAQ
2. Topics: 'Prayer Request' → Collect request (anonymously if preferred) → Power Automate → Email prayer team → Send confirmation with encouraging message
3. Topics: 'Find a Ministry' → Ask about interests → Recommend matching ministry → Provide contact info
4. Topics: 'Daily Devotional' → Return today's devotional from SharePoint library
5. Instructions: 'You are a compassionate, faith-centered assistant. Respond with warmth, scriptural encouragement, and genuine care. Always point people toward community and relationship.'
6. Publish: Website embed + Teams for ministry team use

This is how to structure your day so Copilot agents handle the repetitive work and you focus on what only you can do.

MORNING ROUTINE (6:00 AM – 8:00 AM)

6:00 AM — Business Chat Morning Briefing
How to set it up:
1. Open Teams → Click Copilot (sparkle icon)
2. Create a saved prompt (click bookmark icon):
'Good morning. Create my daily briefing: (1) All meetings today with attendee names and context. (2) Top 5 unread emails from the last 8 hours by priority. (3) Any urgent items needing attention. (4) One verse from Proverbs relevant to leading with wisdom today.'
3. Each morning, open this saved prompt and run it — takes 30 seconds
4. Advanced: Use Power Automate to trigger this automatically and email the briefing to you at 6am

6:30 AM — Email Triage Agent (Power Automate)
Setup (one time, runs automatically forever after):
1. Power Automate → New flow → Trigger: 'When a new email arrives' in your inbox
2. Add condition: If received between 10 PM and 6 AM (overnight emails)
3. Azure OpenAI action: Classify email (Client Inquiry / Invoice / Newsletter / Action Required)
4. Apply logic:
- Client Inquiry → Create draft reply → Flag for your review
- Invoice → Forward to accounting
- Newsletter → Move to Newsletter folder
- Action Required → Create task in Microsoft To Do + Teams notification
5. Turn on the flow — your overnight emails are pre-processed before you open Outlook

7:00 AM — Content Publishing Agent
Setup (weekly, on Mondays):
1. Create a SharePoint list: 'Weekly Content Calendar' (columns: Day, Platform, Content, Hashtags, Status)
2. Power Automate flow triggered on Mondays:
- Call Azure OpenAI with your content pillars and brand voice
- Generate 5 social posts (one per day)
- Add each post to the SharePoint content calendar list
- Send yourself a Teams notification to review
3. Review takes 10 minutes
4. After approval: Second Power Automate flow publishes via Buffer API

WORKDAY ROUTINE (9:00 AM – 5:00 PM)

9:00 AM — Meeting Prep (Before Every Meeting)
1. Open Teams → Copilot
2. Prompt: 'My next meeting is [meeting name] in 20 minutes with [attendees]. Search my recent emails, Teams messages, and shared files for context about this meeting and these people. Give me a 5-bullet prep brief and 2 questions I should be ready to answer.'
3. Read the briefing while having your morning coffee

DURING MEETINGS — Teams Copilot Agent
1. Start every Teams meeting → click '...' → 'Start transcription'
2. Copilot is now active during the meeting
3. Use these real-time prompts in the Copilot pane:
- 'Summarize what has been discussed so far'
- 'What action items have been assigned?'
- 'Are there any open questions?'

AFTER MEETINGS — Teams Copilot Recap
1. Meeting ends → Click 'Recap' in the meeting chat
2. Get: Summary, action items, decisions, speakers
3. Prompt: 'Draft a follow-up email to all attendees with the meeting summary, decisions made, and assigned action items with due dates. Professional but concise.'
4. Send the email from the draft

12:00 PM — Document Creation Agent (Copilot in Word)
For any document you need to create:
1. Open Word → Click Copilot icon → 'Draft with Copilot'
2. Use this template prompt:
'Create a [document type] for [purpose]. Include: [section 1], [section 2], [section 3]. Audience: [who will read it]. Tone: [formal/casual]. Reference our brand as [company name].'
3. First draft ready in under 2 minutes
4. Refine with follow-up prompts

3:00 PM — Data Analysis Agent (Copilot in Excel)
For any reporting or data work:
1. Open your data in Excel → Click Copilot
2. Use this prompt: 'Analyze this data. Identify: (1) Top 3 trends. (2) Any anomalies or outliers. (3) Performance vs. last [period]. Create a chart showing [metric] over time. Summarize findings in 3 bullets.'
3. Full analysis in under 5 minutes

EVENING ROUTINE (5:00 PM – 7:00 PM)

5:00 PM — Business Intelligence Report (Power Automate, automated)
Setup (one time):
1. Power Automate scheduled flow: Weekdays at 5 PM
2. Pull: Sales data from CRM, email stats, web analytics
3. Azure OpenAI: Generate daily performance narrative
4. Send to yourself as Teams message or email
5. Takes 30 seconds to read — replaces 45 minutes of manual reporting

6:00 PM — Tomorrow Prep (Business Chat)
Prompt: 'Review my calendar for tomorrow. For each meeting: (1) Who is attending. (2) What emails or documents are related. (3) What I need to prepare or decide. Give me a simple prioritized action list for tonight.'

TIME SAVED SUMMARY:
Morning email triage: 25 min/day saved
Meeting prep: 20 min/meeting saved
Document first drafts: 45 min/document saved
Data analysis: 40 min/report saved
Daily BI report: 45 min/day saved
Content creation: 90 min/week saved
TOTAL: 3–5 hours saved per day