You started with twelve files and a customer you knew nothing about. Ten tasks later, you have built a package a real account team could take into a real review — and, more to the point, you have practised a method that outlasts the scenario.
What you built
| Deliverable | Written for |
|---|---|
| Account brief | Your own understanding, and your manager's |
| Executive brief | Miriam Graham, who signs |
| Competitive brief | Lidia Holloway, who asked twice and never got one |
| Briefing deck | The three of them, in fifteen minutes |
| Branch analysis | Nestor Wilke, who cannot act on a national average |
| Stakeholder update | Lynne Robbins, read on a phone between meetings |
| Customer proposal | The March review, and whatever procurement does to it |
| Meeting follow-up | Everyone who was in the room, and one person who was not |
| Options decision note | Lynne Robbins, who owes the customer an answer in two weeks |
| Outreach plan and email | Miriam Graham, who has given you one more conversation |
Every one of them was aimed at a named person with a stated concern. That was the difference between a document that gets read and one that gets filed.
What you can now do
You have done each of these against real evidence, not as a demonstration:
- Turn a folder of raw account material into a brief a manager can act on, with every claim traceable to its source.
- Take a year of performance data from a national average down to the one site that explains it — and prove the obvious explanation wrong when it is.
- Build a competitive position that holds up when someone argues the other side of it.
- Write the same findings three ways for three readers without letting a single fact move.
- Audit your own document for figures that drifted, and for claims the sources never supported.
- Turn a recorded meeting into a follow-up the other side would recognize as accurate.
- Turn a customer objection into three costed options, each carrying the assumption it rests on.
- Let the evidence in the files choose what to raise next, and open with the person who signs.
The four moves
Every prompt in this scenario made the same four moves, in the same order:
Goal — what you want produced. Context — why, and who reads it. Sources — what the answer may be built from. Expectations — format, length, and tone.
You saw what happens when one is missing. No goal returns a summary instead of a deliverable. No context aims the tone at nobody. No sources lets Copilot fill the gap from general knowledge. No expectations gives you three pages where you needed one.
Three habits you picked up
Grounding every claim, and labelling the rest. Task 1 had you name a source file under each statement and put everything else under Assumptions. You saw why it mattered: the Calgary root cause was never established, and the 30% at-risk figure came from an offhand remark. A proposal that presents either as fact falls apart the moment someone checks.
Testing the obvious explanation before repeating it. In Task 5 the natural assumption was that the busiest site struggles most. The data said otherwise: Mississauga carried nearly twice Calgary's volume at the best on-time rate in the network. One prompt was the difference between an insight and a plausible-sounding mistake.
Reading your own work as the person who will challenge it. Task 3 had you argue Isaiah Langer's case using your own performance data. Task 7 had you review the proposal as Lidia Holloway. Task 8 had you check a follow-up against a transcript three of its recipients had sat through. Each one surfaced the weakness while there was still time to fix it.
What Copilot did, and what you did
Copilot summarised twelve months of data, restructured drafts for different readers, and listed its own claims back so you could audit them. It did all of that faster than you could have, and without complaint.
What it did not do: decide that Calgary was the story. Refuse to name a cause it could not support. Judge that the recommendation belonged in the first two sentences rather than the last. Notice that the customer's stated requirement mattered more than the metric that flattered us.
Every judgement that decided the outcome was yours. What Copilot removed was the work that was only ever in the way — which is the whole point, and worth remembering the next time someone asks what it is for.
Two things to take with you
The client asked that everyone leave this session with something reusable, not just something completed. These are those two things.
The prompt shape. Every prompt in this lab is the same four moves in the same order. Keep this card:
| Move | What it answers | What goes wrong without it |
|---|---|---|
| Goal | What do I want, in one sentence? | You get a summary when you wanted a decision |
| Context | Who is this for, and what do they already know? | The tone is wrong for the reader |
| Sources | What may it use, and what may it not? | It fills the gaps with plausible invention |
| Expectations | What shape, what length, what must survive? | You rewrite the output instead of the prompt |
The verification checklist. Run it before anything leaves your hands:
- Every figure traced back to the file it came from.
- Every claim either sourced, or labelled as an assumption.
- Units and periods still attached to every number.
- Nothing asserted as fact that the sources only suggest.
- No customer or colleague named in a document that will travel further than they expect.
- The output read once as the person who will challenge it.
Take it to a real account this week
The scenario is finished; the shape it taught you is not tied to it. Pick a live account, gather whatever you actually have — notes, an email thread, a spreadsheet — and run the same first prompt against it.
Build a one-page account brief for <customer name> that my manager can act on before <the next meeting or decision point>.
I am the account executive for this customer. <One or two sentences on what they buy from us, and what has changed recently.>
Use only the files I have attached. Do not add outside information about the company.
Keep it to one page, under the headings Account context, Business drivers, Opportunity themes, and Risks to validate. Name the source file under every claim, and put anything the files do not directly support under a final heading called Assumptions.
If the brief that comes back is thin, the usual cause is the sources line: Copilot can only be as grounded as the files you attach.
Important
Before you reuse anything from this lab with a real customer: Tailwind Traders, the three competing carriers, the contacts, and every rate in the service catalogue are fictional, invented for this exercise. Take the method with you. Leave the numbers behind.