Seat counts are a vendor's metric, not an operator's. Plenty of licences sit in a tenant like unopened kit in a comms room — paid for, provisioned, assigned to someone who logged in once during the pilot and never came back. So when Microsoft reports that Microsoft 365 Copilot has passed 30 million paid seats with net adds more than doubling quarter over quarter, the right reaction isn't scepticism about the number. The number is almost certainly correct. It measures Microsoft's quarter, not your outcomes, and the two have never been the same thing.
What's more interesting is the framing around it. The post is titled "the next measure of AI momentum is work transformed" — which is Microsoft conceding, in a rather elegant way, that seats sold has run its course as a proof point. Microsoft's own proposed measures are depth of use, measurable change in the work itself, the growth and governance of agents, and the redesign of roles and workflows. That concession is worth paying attention to, because it lands right about the time a lot of us are approaching the second or third renewal on Copilot licences bought in the initial enthusiasm.
What a seat count actually measures
A paid seat measures a procurement decision, and procurement decisions in this space have been made for a range of reasons, not all of them about productivity. Some organisations bought Copilot because the board asked what the AI strategy was. Some bought it because it was bundled into a broader agreement negotiation and the marginal cost looked manageable against the discount elsewhere. Some bought a block of seats to run a genuine pilot and never right-sized afterwards, because reclaiming licences is nobody's favourite Tuesday.
None of that makes the 30 million figure dishonest. It makes it orthogonal. The gap between assigned and active is the oldest problem in enterprise software licensing, and Copilot has a particularly wide gap because the skill required to get value from it is unevenly distributed and largely untaught. Two people with identical licences and identical roles can get wildly different results — one treating it as a drafting partner with context, the other as a search box that lies occasionally.
Why 'work transformed' is a harder metric than it sounds
I'm sympathetic to the shift in framing, but let's be honest about the measurement problem. Transformed work is a counterfactual claim: this thing took less time, or produced a better outcome, than it would have without the tool. Counterfactuals in knowledge work are notoriously slippery. If a project manager's status reports now take twenty minutes instead of an hour, did the tool save forty minutes, or did the report get shorter because the tool made it easier to be terse? Both are fine outcomes. Only one is a productivity gain in any sense a CFO recognises.
The trap is that this ambiguity cuts both ways, and vendors and internal champions alike will happily fill it with self-reported time savings from a survey. Self-reported time savings are the softest currency in any benefits-realisation review. If your Copilot business case is denominated in "users report saving 45 minutes per week", multiply that by your fully-loaded hourly rate at your peril. Nobody hands that time back.
The prerequisite nobody puts on the invoice
The operational lesson I keep returning to is that Copilot doesn't have a productivity problem in most tenants. It has a permissions problem. The tool respects the access model you already have, which means it faithfully surfaces every document that HR dropped into a SharePoint site with a broken inheritance chain in 2019, every "anyone with the link" share from a project that finished three reorganisations ago, every Teams file library that got opened up because someone couldn't be bothered raising a request.
Before Copilot, that sprawl was theoretically discoverable and practically invisible. Search was bad enough that nobody found the salary bands by accident. Copilot is good at finding things, which converts a latent governance debt into an immediate incident. I don't think this is a reason not to deploy — I think it's a reason to sequence properly. The oversharing assessment, sensitivity labelling, and a serious look at your SharePoint site lifecycle come before the seats, not after the first awkward discovery.
The uncomfortable arithmetic is that this work is unfunded. The licence cost sits on a line item; the six months of permissions remediation sits on your team's existing capacity. That's the real cost of Copilot, and it doesn't appear in anyone's momentum figures.
What I'd actually instrument
If you want a defensible position at renewal, gather this over a full quarter rather than a fortnight of enthusiasm:
- Assigned versus monthly-active versus weekly-active seats, broken down by department — the departmental split is where the reclamation opportunities hide.
- Surface mix: is usage concentrated in one app? Chat-only usage often means people are using it as a general assistant they could get elsewhere, which is a different value proposition than in-context Word or Excel work.
- Two or three named workflows with a before-and-after measured in something objective — cycle time on a change record, days to close a tender response, first-pass approval rate on a document.
- Support ticket volume attributable to Copilot, including the "why can I see this" and "why can't I see this" categories, which tell you about your permissions estate for free.
- Reclaimed and reassigned seats as a running total, because a licence pool you actively manage is worth more than one you renew wholesale.
The renewal conversation is the real event
At list price, a thousand seats is a meaningful annual commitment on top of your existing E5 estate. The first renewal is where the seat-count era ends and the evidence era begins, and the organisations that will negotiate well are the ones with departmental usage data and two credible workflow case studies — not the ones with a satisfaction survey.
I'd also plan for the licence pool to shrink before it grows. Right-sizing after a broad initial rollout isn't a failure of the programme; it's what a mature deployment looks like. The finance conversation goes considerably better if you frame it that way up front rather than defending it later.
Where I actually land
The momentum is real and the direction of travel is clear enough that I'd not advise anyone to sit this one out. But adopt Microsoft's new framing rather than its old one — and then apply it more rigorously to yourself than the marketing does. Work transformed is the right question. The answer has to come from your own telemetry, your own workflows, and a willingness to say that some cohorts didn't need the licence. That last bit is the part that earns credibility for everything else you ask for.