En bref. Le taux d'adoption de Microsoft 365 Copilot tourne autour de 35,8 % : sur dix licences achetées, environ six dorment. À 28,50 $ CA par personne par mois, c'est une dépense annuelle qui se reconduit toute seule. Cet article montre comment mesurer l'usage réel avec le rapport du centre d'administration, comment licencier par rôle plutôt que par effectif, et pourquoi un développeur n'a probablement pas besoin de Copilot 365.
The number that should decide your licence count
Microsoft 365 Copilot's workplace adoption rate sits at 35.8% — fewer than four in ten people with a licence actually use it. First-year active adoption typically lands between 30% and 55% of purchased seats. A harsher reading puts adoption under 4.5% after three years, with 1% using it weekly.
Treat those figures as a range rather than a verdict. Even the most generous one says the same thing: most organisations are paying for Copilot licences nobody opens.
At CA$28.50 per person per month, twenty-five licences cost CA$8,550 a year — on top of the Microsoft 365 subscription underneath them. If fourteen of those twenty-five never open it, roughly CA$4,800 a year is buying nothing. And it renews automatically, which is the part that makes it permanent.
This is not an argument against Copilot. It is an argument against buying it by headcount.
Why unused seats happen
Blaming users is the wrong diagnosis, and it produces the wrong fix — usually another round of training nobody attends.
Three causes account for most of it.
No task the person was meant to change. A licence arrives with an announcement and no instruction. "Use AI" is not a task. "Draft the first version of your client update in Word, then edit it" is. People who are given a specific job to do differently, in a tool they already open daily, keep going. People given a capability do not.
Weak data hygiene, so the answers are poor. Copilot summarises what it can find. If documents have vague names, live in six places, and there is no source of truth, the answers are confidently wrong. Two bad answers in week one and a reasonable person stops asking. No amount of prompt training fixes a content problem.
No follow-up after launch week. Copilot is a habit product and habits decay. Where leadership uses it visibly and ties it to real work, teams follow. Where the rollout was an email, it dies quietly and shows up later as a renewal invoice.
Measure before you renew
The Microsoft 365 admin centre has a Copilot usage report. It shows who has a licence, who used Copilot in each app (Word, Excel, Outlook, Teams, PowerPoint), and when they last did — over 30, 90 or 180 days.
Use the 180-day window. Thirty days catches a holiday and tells you someone is inactive when they were away.
Three groups come out of it:
Regular users. Active in at least one app most weeks. Keep the licence and find out what they are doing with it, because that is your internal case study for everyone else.
Tried it, stopped. Activity early, nothing since. These are the recoverable ones and they are worth a conversation before a reclaim — usually one of the three causes above, and usually fixable.
Never opened it. No activity in six months. This is the reclaim list.
A caution worth stating: the report tells you a licence was used, not that it was useful. Someone who opens Copilot in Outlook weekly might be dismissing a prompt. Treat the data as a filter for where to ask questions, not as an answer on its own.
License by role, not by headcount
Here is where most of the money goes, and it is a product confusion rather than a waste.
Microsoft sells several different things called Copilot. They are not tiers of one product and they are not interchangeable:
- Microsoft 365 Copilot — works in Word, Excel, PowerPoint, Outlook and Teams, grounded in your company's data through Microsoft Graph. CA$28.50 per user per month, annual, on top of a qualifying Microsoft 365 licence.
- GitHub Copilot — writes and reviews code, in the editor. Around US$19 per user per month for Business, and it does not require a Microsoft 365 licence underneath it.
- Copilot Studio — builds custom agents. Priced in credits, not seats.
- Security Copilot — for security operations, priced separately again.
A developer given Microsoft 365 Copilot has been given a tool for writing documents and summarising meetings. It will not help them write code. GitHub Copilot will — and it is a different purchase, billed separately, which is why it rarely appears in the same conversation as the Copilot 365 rollout.
The common failure is a blanket Copilot 365 purchase for the whole company, including an engineering team that would have gotten far more from GitHub Copilot, and which now has a licence they do not open.
Match the tool to the work:
| The work | The tool |
|---|---|
| Documents, email, spreadsheets, meetings | Microsoft 365 Copilot |
| Writing and reviewing code | GitHub Copilot |
| A custom agent for a specific process | Copilot Studio |
| Security operations | Security Copilot |
Someone who does two of these needs two licences. Someone who does none of them needs none — and "everyone gets one" guarantees a population of the third kind.
The sequence that works
1. Pull the 180-day usage report before the renewal date, not after it. Renewal is when the count can change without a negotiation.
2. Sort people by what they actually do, not by department. Job titles are a poor proxy; a "marketing manager" who lives in Excel and a "marketing manager" who lives in the CMS need different things.
3. Talk to the tried-it-and-stopped group. They are the cheapest adoption you will ever buy, and they tell you which of the three causes is yours.
4. Reclaim the never-opened licences at renewal. Not mid-term — you will usually pay for the term anyway.
5. Keep a small buffer. Reclaiming everything and then fighting procurement for one licence six weeks later is a bad trade. Hold a few.
6. Put the review in the calendar. Annually, two months before renewal. Without a date, this becomes a one-off cleanup and the count drifts back up.
What right-sizing is not
It is not a reason to skip Copilot. The organisations getting value from it are real, and the gap between them and the rest is mostly whether anyone did the work of matching the tool to a task.
It is also not a one-time exercise. People change roles, teams grow, and a licence count that was right in March is wrong by the following March.
And it is not a substitute for readiness. If the tenant has an oversharing problem, right-sizing the licence count means fewer people encountering it — which is a smaller problem, not a solved one.
In short
Microsoft's own adoption figures are the strongest argument against buying Copilot by headcount, and they are more persuasive than anything a consultant can assert. Somewhere between 45% and 65% of purchased seats are not in regular use.
Pull the report. License by role. Reclaim at renewal. And know which Copilot the person in front of you actually needs, because there are four of them and only one of them writes code.
Want help putting this into practice?
We work alongside your team to design, build, and operate the controls described above.