Copilot Pilot Program: Who Goes In, How Long, and How to Skip Pilot Purgatory
The short answer: Microsoft publishes no pilot size, no pilot length and no pass mark for Copilot (checked on its official pages in November 2026). So here's my rule instead. Your AI champions and your leadership team get licences first. Approval takes days. And there is no go/no-go gate.
Everyone else starts on free Copilot Chat and asks for a licence when they hit a real limit. That's how a pilot grows without turning into a committee (committees are not known for speed).
What Microsoft says about a Copilot pilot (and what it leaves blank)
Microsoft gives you the outline of a pilot and leaves the numbers blank. Its current adoption pages talk about a "first cohort" rather than a pilot. Here's what they give you, and what they don't.
| Microsoft gives you | Microsoft leaves blank |
|---|---|
| A rollout team: an executive sponsor, security, governance and an Adoption Champion. | How many people the first group should have. |
| A first cohort: a focused group working on low-risk, high-value scenarios. | How long it should run. |
| A security group in Microsoft Entra ID for that cohort, then you assign the licences to the group. | What number counts as success. |
| Fix oversharing first, and check that Microsoft Purview Audit is on. | How many champions per employee, or how many hours they get. |
| Track results, then decide what to expand, fix or drop. | Whether executives belong in the first group. |
2 warnings if someone is selling you a plan. First, Microsoft's deployment blueprint is a data security guide (oversharing, guardrails, regulation), not a pilot plan. Second, the big rollout story on Microsoft's site is Microsoft's own. It went in phases and ended at more than 300,000 people, which tells you what Microsoft did, not what it recommends to you.
So when a vendor quotes you "Microsoft's recommended pilot size", ask where they found it. Microsoft doesn't publish one.
Who goes in first: my rule for a Copilot pilot program
This part is my recommendation, not Microsoft's. Leadership and AI champions get the paid licences first, and here's why.
1Leadership
They set the example, and they judge the requests that follow. A manager who doesn't use the paid version can't tell whose request makes sense (it's like grading a restaurant you've never eaten at). Microsoft's own checklist starts with an executive sponsor who explains to employees why the company is rolling Copilot out. It doesn't say whether executives get licences first, so that part is my call.
2AI champions
The people who help colleagues get started need the full version to show what it can do (you can't demo a feature you don't have). Give each one a one-page charter, which is also what Microsoft's champion guidance suggests: purpose, boundaries, escalation and expected time commitment. Microsoft gives no hours and no ratio, so you set them.
Draft a one-page charter for an AI champion at our company. Include purpose, boundaries, when to escalate to IT, and how much time it takes each week. Keep it short enough to read in a minute.
31 owner
Name a person. It's usually IT (the CTO or CIO) together with HR, though it varies with which executives actually want to lead it. Whoever owns it decides, so make it someone, not a committee. More on that in my page on executive sponsorship for Copilot.
How many people is that? As many as your champions and your leadership team add up to. Microsoft names no number and I won't invent one. Seats aren't the constraint either: Microsoft removed the old 300-seat minimum in January 2024.
How long should a Copilot pilot run? Until it stops being one
My version has no end date, because it has no gate. A gate ("we expand if the pilot hits X") needs an X, and Microsoft publishes none. In my view, a number invented on day one turns into a reason to wait.
Waiting is the problem, in my view. Gartner's survey of 187 IT leaders found that of those who had finished pilots, only 5% said they were moving to a larger deployment in 2025. (Mind the date. It was published in June 2025, so it's a dated snapshot, not today's rate, and it doesn't say why those pilots didn't move.)
Picture someone in finance who needs Copilot in Excel and is waiting on a meeting (and the meeting, naturally, is not this week). Copilot in Excel is money on the floor, so I want that licence approved in days, not weeks or months.
What I do keep is a review day. The Copilot Dashboard shows a rolling 28 days, so a monthly look matches the data. A review isn't a gate, because nobody else waits for it.
And don't judge anyone in their first week (the Dashboard hasn't even noticed them yet). Data takes up to 7 days to start after a licence is assigned, and the Dashboard runs up to 6 days behind.
What to measure in a Copilot pilot: 4 things
Microsoft names activation, active usage and agent usage as things to track, and gives no targets. These are the 4 I'd look at.
| Measure | Where to read it | The catch |
|---|---|---|
| Active users against licences | Copilot Dashboard, in the Viva Insights web app | "Active" means at least 1 Copilot activity in the previous 28 days. That's a low bar. |
| A harder active rate | Microsoft 365 admin center: Reports, Usage, Microsoft Copilot, Copilot | It counts an intentional action, such as sending a prompt. Opening the Copilot pane doesn't count. |
| 1 finished workflow per champion | Your champions, in their own words | The number is theirs, not Microsoft's. Pick the task before you start. |
| What people say | Talking to them, or the Dashboard's survey sentiment | The reduced Dashboard under 50 licences leaves survey sentiment out. |
What "expand" looks like (without a rollout party)
Nobody else gets pushed into a licence. They use free Copilot Chat properly first. When a real task hits a limit (Copilot on their own files and meetings, Copilot inside Word, Excel and PowerPoint on real documents, or agents on company data), they ask. Their direct manager confirms quickly and IT assigns. The details are in my Copilot licence allocation method.
Already handed out licences in bulk and watching them sit unused? That's the opposite problem, and the 4-week Copilot relaunch covers it. And when the board asks how it's going, here's a one-page Copilot board update.
Questions about running a Copilot pilot
How big should a Copilot pilot be?
Microsoft doesn't say. It asks for a focused first cohort. My answer: as big as your AI champions and leadership team together, because those are the people who go in first.
How long should a Copilot pilot last?
Microsoft gives no duration. In my version it has no end date and no gate, only a monthly review day (the Copilot Dashboard shows a rolling 28 days, so monthly fits).
What is a good success metric for a Copilot pilot?
Microsoft names activation, active usage and agent usage but publishes no targets. I'd read active users against licences in the Dashboard, a harder active rate in the admin center, and 1 finished workflow per champion.
Is there a minimum number of Copilot licences?
Not for the enterprise add-on. Microsoft removed the 300-seat minimum in January 2024, and I couldn't find a current minimum on its pages. For Business plans, 300 users is the maximum.
Why is our Copilot adoption stalling after the initial rollout?
My view: licences went out in bulk with no owner and no champions. The fix is to name an owner, put champions and leadership on paid licences, and let everyone else request one after real use of Copilot Chat.
Is your pilot stuck, or has usage gone quiet?
My Second Wave is a Copilot adoption workshop for rollouts that stalled: 4 sessions of 2 hours over 4 weeks, for up to 20 people. It starts with a licence audit.