Before you start: which Copilot do you have?
"Copilot" is actually three different products. Knowing which one you have decides which parts of this course apply to you.
FREE COPILOT CHAT
Comes with Windows, Edge, and basic M365. Can chat, search the web. Cannot read your documents or emails.
COPILOT PRO
~$20/month personal add-on. Adds Copilot inside Word, Excel, PowerPoint, Outlook for individuals. Critical gap: no Microsoft Graph access, no organisational data, no Teams meeting summaries. Personal use only.
M365 COPILOT
Enterprise add-on (~$30/user/month). Full organisational grounding via Microsoft Graph (emails, files, Teams, calendars), Agents, Cowork, Researcher. This is the only tier that reads your org's data.
Know Your Copilot Surfaces
Copilot lives in four different places. Picking the right one is the single biggest factor in the quality of what you get back.
If you remember one thing from this module, make it this: the wrong surface can't be rescued by a better prompt.
When people first try Copilot, they usually open the web-based Copilot Chat and ask it to do a task like "summarise my sales spreadsheet" or "draft a reply to the email from finance." The output is vague, generic, or just flat out wrong. The conclusion most people reach is "Copilot isn't that good."
But the problem isn't Copilot. The problem is that Copilot Chat in a browser tab cannot see your spreadsheet. It cannot see your email. It cannot see your Teams messages, your calendar, your OneDrive, or any of the data that would make its answer actually useful to you.
There are four different places Copilot lives, and each one has different access to your data. Picking the right surface for your task is the single biggest factor in whether the output feels like a helpful colleague or a random chatbot.
Each surface solves a different problem. They share the same underlying AI but have very different access to your data.
Copilot Chat
The general assistant. Available in the browser (copilot.microsoft.com), in Teams, and via the M365 app. It has two modes: Web mode (searches the public internet) and Work mode (searches your tenant via Microsoft Graph). The free version cannot see your documents at all; the paid version can, but only in Work mode.
In-App Copilot Paid only
Copilot embedded inside each Office app. It reads the document you're actively in and can edit it directly. This is where the day-to-day time savings actually live. Different Office apps have different in-app Copilot capabilities tuned to what that app does: Word is strongest at drafting and rewriting, Excel is strongest at formulas and charts, PowerPoint is strongest at generating decks from briefs.
Agent Mode Paid only
Agent Mode is how Copilot handles multi-step work inside an Office file. Instead of doing one edit per prompt, it plans the whole sequence, makes the edits, checks whether they actually worked, and self-corrects until the result matches what you asked for. Think of it as upgrading Copilot from "answer a question" to "finish a job."
Cowork Beta · Enterprise
Cowork is Copilot's cross-app feature. Agent Mode works inside one file, Cowork works across your whole M365 environment: drafting emails, scheduling meetings, creating documents, posting in Teams, all in one workflow. Each action requires your approval before it executes. Currently in limited beta, so availability varies by tenant and region.
Even within Copilot Chat itself, there are different response modes that dramatically change output quality. Most people never touch them.
| Mode | What it does | When to use it |
|---|---|---|
| Quick Response | The default. Fast, lightweight answers. Powered by the smaller, faster model. | Simple factual questions, quick drafts, casual back-and-forth. |
| Think Deeper | Uses the larger reasoning model (GPT-5.2). Slower, but noticeably better at multi-step reasoning, analysis, and nuance. | Strategy documents, complex emails, comparing options, anything where quality matters more than speed. |
| Researcher | Agentic research mode. Runs a full research cycle, pulls from multiple sources, produces a structured Word/PPT/PDF deliverable. | Board papers, market analyses, policy briefs, anything where you'd otherwise spend half a day Googling and writing. |
| Pages / Notebooks | Persistent workspaces with uploaded files, notes, and context. Copilot grounds every response on that curated material. | Recurring work where the same references apply every time: weekly reports, project decks, standing briefs. |
The single biggest upgrade anyone can make
Switch to Think Deeper for anything more complex than a quick lookup. The quality jump is significant and costs you nothing extra on the same licence. Most people default to Quick Response out of habit and never realise they're asking a faster but weaker model to do serious work.
Diagnosing bad output almost always lands on one of these.
Using Copilot Chat for work that needs the in-app version
You paste a chunk of an Excel file into Copilot Chat and ask it to analyse. It gives you a generic answer because it's working from a text snapshot, not the live spreadsheet with formulas, formatting, and structure. Close Chat, open Excel, use Copilot in Excel. It reads the sheet directly, sees everything, answers accurately.
Staying in Web mode when asking organisational questions
You ask "what did we decide about the facilities renewal?" in Web mode. Copilot has no idea because your organisation's decisions aren't on Bing. Flip the toggle to Work mode and now it can search your emails, Teams messages, and SharePoint for the actual context and give you a real answer.
Using Quick Response for a task that needs Think Deeper
You ask for a strategic review, a complex email, or a comparison of three options. Quick Response gives you surface-level output. Switch to Think Deeper and you'll often get a qualitatively different answer: more structure, better reasoning, fewer obvious gaps. Speed matters less than you think when the output needs to be usable.
A simple decision guide for everyday tasks. When in doubt, come back to this.
If your task is…
Before the quiz, remember
- Four surfaces: Copilot Chat, In-App Copilot, Agent Mode, Cowork. Each has different access to your data.
- Copilot Chat has two modes: Web (public internet) and Work (your M365 tenant). Switch to Work for anything about your organisation.
- Copilot Chat also has response modes: Quick Response (fast), Think Deeper (better reasoning), Researcher (full reports).
- The wrong surface can't be fixed by a better prompt. If Copilot can't see your data, the best prompt in the world produces generic output.
- When working on a document, always use the in-app Copilot for that app, not Chat. It reads the file directly.
Test yourself
Five quick scenarios to check your understanding. Click an answer to see feedback.
Nice work!
You scored 0/5. Mark this module complete below to save your progress.
Write Prompts That Get Usable Output
Most of the difference between "wow" and "useless" comes from how you ask, not what you ask. Here's the pattern.
More than any other single habit, this is the thing that separates Copilot power users from everyone else.
Two people on identical licences, with identical training, can get wildly different output from Copilot. Sometimes 10x different in quality. The difference is almost never which tool they use. It is how they ask.
When you type a vague prompt, Copilot has to guess at what you actually want. It guesses based on statistical patterns from a vast amount of training data. The "average" answer that fits the average meaning of your vague prompt is, by definition, generic. That is what generic output actually is: the AI defaulting to the most common interpretation because you did not give it enough to work with.
When you write a specific, well-structured prompt, Copilot has far less to guess at. It can produce something tailored to your situation, your audience, your constraints. Same model, same licence, same tool. Radically different output.
The good news: the patterns that work are simple and repeatable. The rest of this module gives you those patterns, explains why they work, and shows you common mistakes to avoid.
Same task, three approaches. The difference is not length, it is structure.
Let the AI do the thinking, not you
The "Better" prompt is good, but you had to think through exactly what structure you wanted. The "Best" prompt offloads that thinking onto Copilot. You describe the situation and goal, Copilot identifies the gaps and asks you about them. Clicking answers to Copilot's questions takes seconds. Writing the perfect prompt from scratch takes minutes. This is the single highest-leverage habit to build.
Most great prompts have five ingredients. Missing any one of them usually explains why output feels off.
Five-ingredient prompt, annotated
Each colour shows what that part of the prompt does.
Role
Who Copilot is being in this response. Changes tone and perspective.
Task
What you actually want produced. Be specific.
Context
Who it is for, what came before, the situation around the task.
Constraints
Length, tone, what to avoid, hard rules.
Format
The shape of the output: sections, options, structure.
You do not need all five every time. For quick questions, "task" alone is fine. But for anything you would call "real work," checking you have covered at least Task + Context + Constraints will transform your hit rate.
Toggle between levels to see how the same Excel request transforms as you add ingredients.
Live prompt builder
Same pivot table request, three quality levels.
If your output is consistently disappointing, check if you are falling into one of these patterns.
Not telling Copilot who it is
Same content. Completely different feedback. A role directs the lens Copilot uses. Without one, you get generic, blandly positive feedback.
Saying "good" and expecting Copilot to know what that means
"Better" means nothing to Copilot. Define what "good" looks like: more direct, warmer, more formal, shorter. Give it a direction, not a vibe.
Giving Copilot a huge task with no scope
Huge open-ended tasks produce huge vague responses. Narrow the scope to one chunk at a time, and Copilot gives you usable output for that chunk.
Writing a 500-word prompt when a question would do
You do not need to guess what context matters. Let Copilot ask. Clicking answers is faster than writing a perfect prompt from scratch.
Starting with "build" when you should start with "plan"
Build prompts lead to one wrong version that you then have to explain and re-explain. Plan prompts lead to the right version first time.
If you remember one thing from this course, make it this.
Stick it at the end of any meaningful prompt. Copilot will pause before writing, identify what it needs to know, and present you with specific clarifying questions, often with clickable options.
You click answers in under a minute. Then Copilot writes the response using your actual input rather than its guess. The quality difference is significant and the effort difference is almost nothing.
Tapping is fast, thinking ahead is slow
Human beings are bad at knowing in advance what questions an AI needs answered. We miss things. Copilot is actually quite good at identifying information gaps in its own context. So let it surface the gaps, and fill them by tapping. You get all the benefit of a detailed prompt without having to write one.
Do not start with "make me a spreadsheet." Use Chat to align on what you actually want. Once it is clear, then build.
Chat turns are cheap. Builds are expensive. A 30-second chat conversation to align on the spec saves you a 30-minute rework cycle on a wrong-shaped spreadsheet. This is the single most important workflow pattern for any task with real consequences.
Eight reusable patterns. Save them, modify them, use them constantly.
The drafter
When you need text written from scratch.
The editor
When you have a draft and need feedback.
The analyst
When you want insights from a file or data.
The summariser
When you want a tight summary.
The comparer
When you need to weigh options.
The explainer
When you want to understand something.
The builder
When you need a real deliverable made.
The reframer
When you are stuck or output is not working.
Before the quiz, remember
- Bad prompts get generic output, not because Copilot is bad, but because "average" is the default when context is missing.
- Five ingredients matter: Role + Task + Context + Constraints + Format. You do not need all five every time, but check you have at least Task + Context + Constraints for real work.
- "Ask me questions before you start" is the single most useful phrase. Stick it on any real prompt.
- Plan first, build last. Cheap chat turns, then expensive build. Never the other way around.
- Define "good": more direct, warmer, shorter, more formal. "Better" means nothing on its own.
Test yourself
Five questions on prompt patterns. Pick the best answer in each case.
Well done!
You scored 0/5. Mark this module complete below.
Ground Copilot in Your Work
Copilot has access to everything in your Microsoft 365 tenant - emails, files, Teams, SharePoint. Most people never turn this on.
This one toggle is the biggest unused feature in Copilot.
Web mode
Copilot searches the public web. Useful for research, but it cannot see anything inside your organisation.
Work mode
Copilot searches your organisation's data via Microsoft Graph - your emails, files, Teams messages, calendar, SharePoint sites. Nothing leaves your tenant.
For any organisation handling sensitive information about customers, employees, or finances, Work mode is a hard requirement - Web mode would leak context to public search. Set Work mode as the default and train staff to switch only when explicitly researching public topics.
Point Copilot at a specific file, person, or site instantly.
Type / in the prompt box
A picker appears showing files, people, and SharePoint sites you have access to.
Select the reference you want
Example: /Q4-budget.xlsx, /Customer-Service-team, /Customer-Data-Policy.
Ask your question
Copilot reads the referenced file or context directly into the conversation. Example: Summarise the key points from /Q4-budget.xlsx and draft an email to /Grant.
Extend Copilot's reach to non-Microsoft tools, and create persistent reference spaces.
Connectors
Settings → Connectors → Browse → Add. Link Slack, Google Drive, Notion, Jira, and 50+ others. Once connected, Copilot can pull from them the same way it pulls from your M365 data.
Notebooks
A saved space to collect files, notes, and references for a recurring topic. Think of it as a project folder Copilot always has open. Ground your prompts on a Notebook so Copilot draws from your curated materials.
Memory (Work IQ)
Copilot learns from your style, preferences, and habits over time. Check your personalisation settings to see - and manage - what it remembers about you.
Everything Copilot reads in Work mode stays inside your Microsoft 365 tenant. It is not used to train Microsoft's models. That said: Copilot can only respect permissions that are correctly set on the underlying files. If a sensitive SharePoint site is set to "Everyone in the org," Copilot will surface it to everyone. Get your SharePoint permissions right before rolling Copilot out broadly.
When you ask a question in Work mode, Copilot can pull from multiple layers of your Microsoft 365 environment. Understanding what it sees helps you ground prompts accurately.
| Layer | What it contains | How Copilot uses it |
|---|---|---|
| Microsoft Graph | Your emails, calendar, files in OneDrive and SharePoint, Teams messages, contacts. | Default source of truth in Work mode. Pulls relevant items based on your prompt. |
| Active document | The specific file you have open in Word, Excel, or PowerPoint. | Read directly by in-app Copilot. Most accurate source when a file is in front of you. |
| Slash references | Specific files, people, or SharePoint sites you select with the / picker. | Pulls the referenced item into this specific conversation as primary context. |
| Connectors | Non-Microsoft tools: Slack, Google Drive, Notion, Jira, and 50+ others. | Searched alongside Microsoft Graph when relevant to your prompt. |
| Notebooks / Pages | Curated collections of files, notes, and references for a recurring topic. | When Copilot is "grounded on" a Notebook, it draws from that material first. |
| Work IQ / memory | What Copilot has learned about your style, preferences, and recurring work patterns. | Shapes tone, format, and defaults across sessions. Manageable via personalisation settings. |
A reliable pattern for any prompt that needs your real organisational context.
Before Copilot goes live for all staff, walk through these four checks. This applies to any organisation handling sensitive customer, employee, or financial data.
SharePoint permissions audit
Run a site-level permission review. Anything set to "Everyone in the org" will be surfaced by Copilot to everyone. Tighten sensitive sites (HR, finance, customer records, financial data) before enabling Copilot broadly.
Sensitivity labels
Apply Microsoft Purview sensitivity labels to confidential content. Copilot respects these labels: content marked "Confidential" or higher is handled differently from open material.
Pilot group first
Enable Copilot for a small pilot group before organisation-wide rollout. Watch what they use it for, what surfaces unexpectedly, and fix the gaps before scaling.
Acceptable-use guidance
Publish a one-page guide for staff: when to use Copilot, when not to, what kinds of content should never be typed into prompts, and who to ask if uncertain.
Copilot does not create permission problems, it surfaces existing ones
If Copilot surfaces a sensitive document to the wrong person, the root cause is almost always that the document was already shared too broadly. The fix is not to disable Copilot, it is to fix the underlying SharePoint permissions. Rolling Copilot out forces the permission hygiene conversation that most orgs have been putting off for years.
Before the quiz, remember
- Work mode vs Web mode: Work = your tenant, Web = public internet. Most people never switch.
- Slash references (/file, /person, /site) ground Copilot on specific items instead of guessing.
- Data stays in your tenant in Work mode and is not used to train Microsoft models.
- Copilot respects existing permissions. It cannot see what you cannot see, but it will surface anything you technically can see. Audit permissions before rollout.
- Notebooks and Connectors extend what Copilot can ground on beyond the default M365 apps.
Test yourself
Five questions on grounding, modes, and data boundaries.
Great work!
You scored 0/5. This is the module IT teams most often ask about.
Build Real Things in Word, Excel, PowerPoint
This is where the time savings actually live. Every Office app has Copilot embedded - and every app has its own patterns worth knowing.
Concrete examples you can try today. All require a paid Copilot licence.
Word
Drafting, rewriting, and editing long-form text. Copilot reads the whole document and can edit in place.
- "Draft a 400-word newsletter based on the bullet points I pasted."
- "Rewrite this paragraph in plain English for a general audience."
- "Add a summary at the top based on the three sections below."
Excel
Formulas, pivot tables, chart suggestions, and data cleaning. Copilot reads the sheet you have open.
- "Suggest a formula to calculate running total by month."
- "Create a pivot table grouping sales by product line and quarter."
- "Highlight rows where budget variance is over 10%."
PowerPoint
Generate decks from a Word doc or outline. Edit slides, change tone, add speaker notes.
- "Create a 10-slide deck from /Q4-strategy.docx."
- "Simplify this slide so it has no more than 5 bullet points."
- "Add speaker notes to each slide in a warm, conversational tone."
Outlook
Summarise long email threads, draft replies, coach your tone before you send.
- "Summarise this thread in three bullet points."
- "Draft a warm but firm reply saying we can't attend."
- "Coach me: is my tone too blunt here?"
Teams
Meeting recaps, action items, catch-up summaries. Auto-generated after any recorded meeting.
- "What did I miss in the first 20 minutes?"
- "List all action items and who owns them."
- "Draft a follow-up email with key decisions."
OneNote / Loop
Restructure messy notes, turn scribbles into plans, collaborate on live pages.
- "Turn my bullet points into a structured project plan."
- "Summarise this page as three key decisions."
- "Create a checklist I can share in Teams."
Everything Copilot creates in Office apps is a real, editable .docx, .xlsx, or .pptx - not a rendered image. Your team can keep working on it normally, version history still works, and it saves to OneDrive or SharePoint automatically.
There is a single pattern that works whether you are in Word, Excel, PowerPoint, Outlook, or Teams.
Every in-app Copilot follows the same general pattern. It reads the document you have open. It accepts a prompt about that document. It produces an edit, a suggestion, or a new piece of content grounded on what is already there.
The mistake most people make is asking Copilot in the wrong app for the task. They open Word and ask for a chart. They open Excel and ask for a long narrative explanation. Each app has its own Copilot optimised for the kind of work that app does.
Pick the app for the job, then use the in-app Copilot there. Resist the urge to do everything from Copilot Chat.
Word is the strongest Copilot surface for any task involving prose.
| Task | Prompt pattern |
|---|---|
| Draft from notes | Draft a [doc type] for [audience] based on the bullets above. Keep it under [N] words, warm but professional. |
| Rewrite for audience | Rewrite this paragraph for [team members / leadership / customers]. Plain English, no jargon. |
| Add a summary | Add a 150-word executive summary at the top, capturing the three key decisions. |
| Fact-check against source | Compare this document to /[source-file]. Flag anywhere we have changed what the original said. |
| Reformat / tighten | Cut this document to 500 words without losing the main points. Use headings and bullets where it helps. |
Excel Copilot is strongest with clean, tabular data. Messy data confuses it, so start by tidying.
| Task | Prompt pattern |
|---|---|
| Formula help | Suggest a formula to [calculation]. Explain what each part does. |
| Pivot table | Create a pivot table showing [rows] by [columns], with [value] and a grand total row. |
| Highlight outliers | Highlight rows where [condition]. Use conditional formatting: red for [case], green for [case]. |
| Data cleaning | This data has inconsistent formats. Fix the dates, split full names into first/last, flag duplicates. |
| Summary analysis | Analyse this data. Give me the three most interesting patterns and two surprising findings, in plain English. |
Clean data in, useful analysis out
Copilot in Excel is only as good as the data structure you give it. Well-named headers, consistent data types in each column, no merged cells, no blank rows. If you do nothing else, make your headers clear and your columns consistent. That alone transforms output quality.
PowerPoint Copilot can build a full deck from a document. Reliable for structure, headlines, and speaker notes. Inconsistent on layout, visual hierarchy, and brand fidelity. Treat the first output as a draft skeleton, not a finished deck.
| Task | Prompt pattern |
|---|---|
| Generate from doc | Create a 10-slide deck from /[briefing-doc]. One key point per slide, short bullets. |
| Slide-by-slide rewrite | Rewrite this slide to have no more than 5 bullets, each under 10 words. |
| Speaker notes | Add speaker notes to every slide: warm, conversational, about 30 seconds of speaking per slide. |
| Consistency pass | Review the deck for consistency. Flag slides where tone, structure, or formatting does not match the rest. |
| Executive summary slide | Add an opening summary slide capturing the three key decisions leadership needs to approve. |
The highest daily time saver for most people. Email is the universal task.
| Task | Prompt pattern |
|---|---|
| Thread summary | Summarise this thread in three bullets: who wants what, decisions made, what is still open. |
| Warm but firm reply | Draft a reply declining the request. Warm tone, acknowledge their effort, explain we cannot this quarter, offer an alternative. |
| Tone coaching | Coach me on this draft. Is my tone too blunt? Too soft? What would you change? |
| Reply with specific context | Reply to this email using the context from /[relevant-doc]. Keep it under 100 words. |
| Scheduling helper | Draft a reply suggesting three 30-minute slots next week that work with my calendar. |
Meeting recaps and catch-ups. Underused because most people do not know it is there. Note: meeting recap only works if transcription was enabled before the meeting.
| Task | Prompt pattern |
|---|---|
| Catch up after joining late | What did I miss in the first 20 minutes? |
| Action items extract | List all action items from this meeting with who owns them and rough deadlines. |
| Decisions log | What decisions were made in this meeting? For each, who made it and what was the rationale? |
| Follow-up draft | Draft a follow-up email to the attendees summarising decisions, actions, and next steps. |
| Sentiment / pushback check | Was there disagreement or hesitation about [topic]? Who pushed back and on what grounds? |
This is a subtle point that makes a big difference to how teams actually use output.
Everything Copilot creates inside Office apps is a real, editable .docx, .xlsx, or .pptx file. Not a rendered image, not a screenshot, not a PDF export. A normal Office file that your team can keep working on using every tool they already know.
Version history still works. Track changes still works. Co-authoring still works. Saves go to OneDrive or SharePoint automatically, which means sharing, permissions, and archival all work the way they already do in your organisation.
The practical upshot: Copilot fits into your existing workflow rather than creating a parallel one. No exporting, no converting, no copy-paste round trips. If your organisation has a document lifecycle already, Copilot respects it.
Before the quiz, remember
- Each app has its own Copilot, optimised for the work that app does. Use the in-app version, not Copilot Chat.
- Word for prose, Excel for data, PowerPoint for structure, Outlook for email, Teams for meetings. Match the app to the task.
- Clean input, clean output. Especially in Excel: consistent headers, clear data types, no merged cells.
- Output is always native files, editable in Office, saved to OneDrive or SharePoint, with version history intact.
- Highlight text first, then ask, when you want Copilot to work on a specific selection inside a larger document.
Test yourself
Five scenarios - match the task to the right app and approach.
Nicely done!
You scored 0/5. You're halfway through the course.
Automate with Agent Mode and Copilot Studio
The shift from "Copilot answers questions" to "Copilot does tasks". This is where real hours get saved.
Three tools that do different things. Pick the right one.
Agent Mode
Inside Word, Excel, PowerPoint. You give it a multi-step task, it plans, edits, checks, and self-corrects until the job is done. Best for one-off jobs that would take 20+ minutes of manual back-and-forth.
- "Turn these 30 bullet points into a polished 5-page Word document with headings and a summary."
- "Analyse this sales data and build the three most useful charts, then write a one-paragraph summary."
- "Refine this 15-slide deck so every slide has a clear point, consistent formatting, and speaker notes."
- "Clean this messy spreadsheet: fix inconsistent dates, split name columns, flag duplicates."
Cowork (Preview)
Copilot's cross-app feature. Currently in limited beta. Each action requires your approval before it executes. Best for workflows that span emails, meetings, documents, and Teams.
- "Draft a reply to this email, schedule a follow-up meeting, and create a Word summary."
- "Find every email about the facilities project, archive them, and create a Teams channel."
- "When a new lead form is submitted, thank them by email and add them to the CRM."
- "Summarise this week's leadership meeting and send the recap to the department heads."
Copilot Studio Agents
Build your own no-code agent that handles a recurring task. Describe the workflow in plain English, test it, publish it to your team. Best for things you (or your team) do over and over.
- A customer service agent that drafts customer-contact notes from rough descriptions.
- A finance agent that checks invoice formats and flags errors before approval.
- An onboarding agent that walks new employees through orientation steps.
- A reporting agent that builds the monthly board pack from standard templates.
Don't try to automate a whole business process on day one. Pick one annoying, repeatable task that takes someone 10-30 minutes. Build an agent for that. Iterate. Most organisations that succeed with Copilot Studio build five small agents before they attempt anything ambitious.
Automation is intoxicating. Before you build anything, make sure the task is actually a good automation candidate.
| Automate when... | Do not automate when... |
|---|---|
| The task is repeatable with the same inputs and outputs each time. | Every instance is subtly different and requires judgement. |
| It takes 10+ minutes per occurrence, at least weekly. | It only happens occasionally and does not take long. |
| The cost of getting it wrong is low or easily caught. | A mistake would damage relationships, miss compliance, or be unrecoverable. |
| The process is already well documented or easily described. | The process lives in someone's head and nobody has written it down yet. |
| Humans approve the output before it goes live. | The output goes straight to external parties with no review. |
Document the process first, automate second
The single biggest predictor of whether an automation will succeed is whether the process it replaces was clearly documented beforehand. If you cannot describe the workflow in plain English to another human, Copilot Studio will not be able to automate it either. Write it out, test it as a checklist a person could follow, then automate.
The core steps for building a simple Copilot Studio agent. No code, just clear description.
Pick a small, annoying task
Examples: formatting new lead confirmations, drafting weekly department updates from standing inputs, thanking customers by tier. Not "run the whole programme", just one recurring chunk.
Describe it in plain English
"When a new lead form is submitted, draft a welcome email that mentions their region and interests, and CC the regional account manager. Send me the draft for approval."
Give it access to relevant data
Point it at the SharePoint list of regional managers, the customer database, your organisational tone guide. The agent grounds its output on those references.
Test with real examples
Run 5-10 real examples through it. Check the output. Tweak the instructions where it misses. Repeat until it is reliable at least 9 times out of 10.
Publish with approval gate
Always require human approval for the first 50 runs. Then loosen gradually if confidence is high. Keep approval for anything that sends externally.
Practical agents that reliably save time across typical business workflows.
Client interaction note drafter
Staff describe a client interaction in two or three lines. The agent drafts a properly structured note: who, when, context, follow-up actions, flag for escalation if needed. Reduces admin burden while keeping records consistent.
Customer thank-you generator
Pulls this month's customer activity from the CRM, drafts personalised thank-you messages by category (regulars, new, milestone purchases), sends to the account manager for approval before dispatch.
Invoice pre-check agent
Before invoices go to approval, the agent checks them against standard formats: PO number, matching approval path, budget code, duplicates. Flags issues and routes clean ones through faster.
Monthly board pack assembler
Gathers standard monthly inputs (sales, headcount, project status, customer metrics) from their source spreadsheets, formats them into the standing board template, drafts the narrative sections for review.
New employee onboarding agent
When a new employee starts, the agent sends the orientation schedule, links policies to acknowledge, sets calendar invites for induction, and chases missing confirmations for up to two weeks.
Before the quiz, remember
- Agent Mode: one-off multi-step jobs inside a single file (Word, Excel, PowerPoint).
- Cowork: cross-app workflows with human approval at each step. Still in beta.
- Copilot Studio Agents: reusable no-code agents for recurring workflows. Published to your team.
- Start small. One annoying 15-minute task, not the whole department. Iterate.
- Document before you automate. If you cannot write the process in plain English, an agent will not be able to follow it either.
- Keep approval gates especially for external-facing output until confidence is established.
Test yourself
Five questions on when to automate, when to agent, and when to leave things alone.
On a roll!
You scored 0/5. Two modules to go.
Power Tips and Stretching Your Licence
Small habits that separate light users from heavy users. Most take 30 seconds to pick up.
Daily habits that compound
Summarise /Q4-sales-report.pptx and draft an email to /John Smith.Feature discoveries most people miss
Most "Copilot is broken today" moments have a specific, fixable cause. Here is the diagnostic order to work through.
Output is generic or vague
Check: are you in the right surface? If you are in Copilot Chat asking about a specific document, move to the in-app Copilot for that document. Then check: have you given it audience, constraints, and format? If not, add them. Then check: have you tried Think Deeper mode?
Copilot cannot find the document you mean
Check: are you in Work mode, not Web mode? Then check: do you have permission on the file you are trying to reference? Then check: is the file in OneDrive or SharePoint, not on your local machine? Local files are invisible to Copilot.
Output references something that never happened
Copilot is hallucinating, usually because it was underfed on context. Ground the prompt on specific files with slash references. Ask "cite the source" to force it to link claims to real data. If it cannot cite, it was guessing.
Tone feels off, or not like you
Give Copilot explicit tone direction. Reference an example of your voice: "Match the tone of /[past-email-or-post]." Your Work IQ / memory also shapes this over time, so using Copilot regularly on your actual writing improves defaults.
Output is technically correct but not useful
You are probably asking the wrong question. Step back and use the reframer pattern: "I have been trying to [goal] and getting stuck. Ask me five questions to understand the real problem, then suggest a different approach." Copilot will surface assumptions you are making.
Tips that live outside day-to-day use but make a big difference to how Copilot lands across an organisation.
Admin dashboard fundamentals
Before the quiz, remember
- Work mode is the single biggest free upgrade. Most staff default to Web mode and never switch.
- Slash references (/) stop Copilot guessing and ground it on specific files, people, or sites.
- Think Deeper for hard tasks, Quick Response for simple ones. Same licence, better output when needed.
- Highlight-then-ask focuses Copilot on a selection rather than a whole document.
- Pin common prompts to halve the friction on recurring tasks.
- Admin: check adoption reports monthly and reallocate unused licences. The data surfaces who needs help.
Test yourself
Five questions on power-user habits and hidden features.
Almost there!
You scored 0/5. One module left.
Go Deeper
Official Microsoft learning, certifications, and community resources. Bookmark what's relevant to your role.
Test yourself
Five questions pulling together everything from the course.
You did it!
You scored 0/5. Mark this module complete to finish the course.
Course complete
You've worked through all seven modules. You now know more about Microsoft 365 Copilot than the vast majority of people who have it installed. Want help rolling this out across your team?
