Meet Copilot
& communicate faster.

Fingers up, 1 to 5: how much of your week goes into writing, searching, or formatting?

Name · department · and one task you'd happily never do by hand again.

Three freedoms.

Questions board · open all day

menti.com/al5kyzvcip3a

Open it now on your phone. Anything that pops into your head goes here.

Scan to join

By 2pm you'll have made Copilot draft, summarise, and dig through documents for you — and you'll know exactly where NOT to trust it.

First: a quick 30-minute check. No grades, no catch.

The Map Test.

Part 1: rate yourself — no right answers, be honest. Parts 2–3: pick the best answer — think it through, don't guess, leave blanks if unsure.

30:00

Start 30:00 +2 min Reset Open the Map Test ↗

Scan to open

Six modules. Tap one to jump.

So what is AI — and what can it actually do?

Software that learns patterns from examples instead of following rules someone wrote by hand. Nobody coded "how to write a polite reply" — it read millions of them.

And what's an LLM?

Large Language Model. Large — trained on an enormous amount of text. Language — words are all it deals in. Model — a pattern machine, not a filing cabinet.

AI is the field. An LLM is the specific kind that reads and writes language — and that's what sits inside Copilot.

Why should you, as an employee, care?

It doesn't replace the expertise. It removes the drag around it — so the expert part gets your day.

Shout the next word.

That reflex — guessing the next likely word — is the whole engine. Let's see how a machine learns to do it.

It remembers the patterns — not the pages.

Like an intern who read almost everything ever written — fluent about the world, yet it doesn't "know" your customer or your policy. Training = learning patterns from examples. Not a search engine. Not a database.

One piece at a time.

Each step picks the most likely next piece given everything so far. Which means a well-set-up prompt literally steers what comes next. That's why prompting works — more on that shortly.

Fluent ≠ correct.

It predicts plausible text — so when it has a gap, it fills it with a confident-sounding guess. And it sounds exactly as sure when it's wrong.

That's the engine — patterns, tokens, occasional confident fiction. Now Microsoft wires it into your apps and points it at your own work.

That's Copilot.

The same engine — wired into your work.

Your files

Your email

Your chats

Your meetings

AI engine+ Microsoft Graph

 Word

 Excel

 PowerPoint

 Outlook

Connected to your permitted work content through Microsoft Graph — your files, emails and chats, only what you can already access.

A brilliant new intern — with a badge.

It still needs briefing, and it still needs sign-off. You own the output.

Same kind of engine — very different boundaries.

• 1Grounded in your organisation's content — not just its training.
• 2Respects permissions — it surfaces only what you could already open.
• 3Enterprise Data Protection — your data stays in the tenant and never trains the model.

Watch the gold dot: it keeps trying to leave — and it can't. That boundary is the whole reason a bank can use it.

The app map — tap a card.

Word

Draft & rewriteToday

Excel

Ask your dataDeep-dive Day 2

PowerPoint

Deck in minutesDeep-dive Day 2

Outlook

Inbox triageToday

Two ways you'll meet it: in-app Copilot — inside the document you're working on — and Copilot Chat, the standalone chat that can pull across your files.

Your first win — pick a starter and run it.

Open E1.1 in your lab — a menu of ready starters is waiting there. Pick whichever appeals, copy it, and run it in Copilot Chat.

Done looks like: a result you'd actually consider using. Finished early? Steer it — shorter, warmer, or for a customer who's already complained twice.

Some results were sharper than others. The difference wasn't the tool — it was how you asked.

That's the next 45 minutes.

Break — back at 10:05.

Tap the ☕ in the nav bar to run the timer.

One volunteer describes a shape. Everyone draws it blind — no questions allowed.

Vague instructions, vague results — and you couldn't even ask follow-ups. Copilot is the same, except you CAN keep talking to it.

CTFT — four ingredients.

You rarely need all four spelled out. But when a result comes back wrong, CTFT is how you find the missing ingredient and fix it.

CContext — the setup.

Treat the AI like someone very smart who started at Burgan today. Broad general knowledge — but it knows nothing about your situation. You fill it in.

TTask — the what.

Exactly what do you want it to do? Don't say "help me" — that's vague. Name the action: write, create, summarise, analyse.

Keep it to one main task per prompt. "Write a memo AND build a deck AND analyse the budget" is three prompts — do them separately for better results.

FFormat — how it's structured.

Bullets or paragraphs? Short summary or long write-up? Formal report or quick email? Say nothing and the AI picks for you — and it may not match what you need.

TTone — how it sounds.

Same facts, very different feel: "Employees must comply with the new policy immediately" versus "We're excited to introduce a policy that gives you more flexibility." Tone sets the relationship.

Think about your audience: who's reading this, and what's your relationship with them? That's what picks the tone. C·T·F·T — four elements, every time.

One terrible prompt. Four rounds. Watch it get visibly better.

Scenario: a customer's replacement debit card has been delayed. One CTFT ingredient per round — run it each time, record every round on your Prompt Recipe Card. Rounds 0–1 together; 2–4 on your own device.

Let's watch it be generic.

Give it the situation.

Say what to do.

Shape it.

Set the voice — and the limits.

Nobody nailed it in one shot — and that's the method.

That's not failure. You steer with follow-ups. The first answer is a first draft.

Build a prompt for your own work.

Open E1.3 in your lab — the Prompt Builder is there. Fill the fields (Role · Context · Task · Format · Tone · Example), assemble, and test it in Copilot. Best one goes on the wall as Prompt of the Day.

Role lives inside Context; Example is optional — this is CTFT made explicit, not a new framework.

A finished one — ingredients highlighted.

Sample data — never paste a real client's figures today.

You can steer Copilot now. Next: point that skill at the thing that eats your morning — documents in Word.

Watch the page — tap a move.

Move 4 is the callback to Copilot’s boundary: refer to a file, so the draft is grounded in real content — the house voice, not a guess.

Bullets in — sendable letter out.

Done looks like: a letter you'd be comfortable sending, grounded in the branch-letter tone — plus a five-point summary of the longer notice. All three assets: /day-1-assets on this site (and the shared folder). Close: one volunteer reads before-bullets → after-letter.

Break — back at 11:30.

Tap the ☕ in the nav bar to run the timer.

Six messages of chaos → one summary.

Turn the tone dial.

Same facts, three voices — you choose per audience. That's the same steering skill from earlier, pointed at email. Fourth move to show live: catch up after time away — the inbox mountain, pre-sorted before you've finished your coffee.

The escalating complaint — tamed.

Then iterate the tone twice: more concise → slightly warmer, less corporate. Done looks like: a summary that correctly separates promised from outstanding, and a reply you'd send.

Lunch — back at 13:00.

Tap the ☕ in the nav bar to run the timer.

In one sentence: the best thing Copilot did for you this morning.

Two or three voices — then straight in.

Long material → short brief.

Brief it, extract it — then verify it.

The verify step is not optional — open the release and check it yourself. Full report + prompts are in E1.5 in your lab.

It summarises beautifully — and still misplaces a number.

Look at the wall.

• ✓The tasks you said you'd never do by hand again — tick off the ones we solved today.
• ✓The Prompt of the Day — photograph it, steal it, reuse it.
• ✓One line each: "the one thing I'll use tomorrow is…"
• ✓Take your Prompt Recipe Card — leave it where Monday-you will find it.
• ✓Tomorrow ends with the POST Map Test — same sheet as this morning, so you see exactly how far you travelled.

Today you made Copilot write for you.

Tomorrow you make it think with your data and build your decks — and we get serious about doing all of it safely in a bank. Same room, same start time. Bring your Recipe Card.

your data

your decks

🛡done safely