A finance lead I know pulled up her Microsoft 365 Copilot renewal a few weeks before it was due. The number was not small. Copilot is priced per seat per month, and her company had bought a lot of seats, because the rollout a year earlier had been enthusiastic: everyone who might benefit got a licence, on the theory that you do not want to be the person who gatekept the productivity tool.
So she asked the obvious question before signing again: how many of these seats are people actually using?
Silence. Not because nobody cared, but because nobody could answer it. She had one number cold, seats purchased, it was on the invoice. The number she actually needed, seats being used, did not exist anywhere she could reach. The admin center showed who had a licence assigned. It did not show who had opened Copilot this month, let alone who had used it enough to justify the cost.
That gap is the whole story of Copilot economics right now, and it is worth understanding whether or not you ever fix it, because almost every team is quietly in her position.
Two numbers, and you can only see the easy one
Every per-seat software rollout has two numbers:
- How many people got a licence. Easy. It is on the bill and in the admin center. This is the number everyone has.
- How many people actually use it. Hard. It requires connecting "who has a seat" to "who has real activity," and the two live in different places, if the activity data is reachable at all.
The trouble is that the decision, renew, cut, or reallocate, depends entirely on the second number, and the second number is the one nobody has. So renewals get made on the first number by default: "we bought 500 seats, renew 500 seats," with no idea whether 500 or 150 of them do anything. That is not a spending decision. That is a spending reflex.
Why this is a FinOps problem, not an IT one
It is tempting to file this under "IT should track that." But it is really the same shape as every cloud cost problem I deal with: you are paying for provisioned capacity, and nobody is checking utilization. An unused Copilot seat is exactly an idle EC2 instance. It is provisioned, it is billing, and it is doing nothing, and it survives because no one connected "we pay for this" to "is anyone using this."
AI seat licences are just the newest place that pattern shows up. The State of FinOps world has been saying it all year: cost management stopped being about the cloud bill and now covers AI and SaaS spend too. Copilot seats are the poster child, per-seat, AI-priced, enthusiastically over-provisioned, and almost never measured against real use.
What you actually want to see
If you are going to make an honest renewal decision, here is the view you need, and it is a good checklist regardless of what tool you use to get it:
- Bought, assigned, used. Three numbers, not one. Seats you pay for, seats handed out, and seats with real activity behind them. The gap between the second and third is your waste.
- Per person. Active days, prompts, and when each person last actually touched Copilot. This is what turns "we have low adoption" into "these specific 80 people have not opened it in a month."
- Adoption over time. Daily active users against licensed seats over the last couple of weeks, so you can see whether usage is growing, flat, or quietly dead.
- App footprint. How many Copilot-enabled apps each person really works across, because a seat used in one app is a different story from one used across many.
- Renewal impact. The one that makes it real: what the unused seats cost you each month, stated as a number you can act on.
That is the difference between renewing on a reflex and renewing on evidence.
Where this connects to what we built
I work on ZopNight, and this is a problem we recently made visible, so I will be straight about it rather than coy: if you connect your Microsoft 365 tenant under Settings then Integrations, the Cost Reports AI Estate view now shows exactly the seat-licence picture above, bought vs assigned vs used, per-person activity, adoption over time, app footprint, and what the unused seats cost per month. Not because you need our tool to care about this, but because the data was genuinely hard to assemble and that is the gap we closed.
The reason I am writing this is not the feature though. It is the mindset. Even if you never touch ZopNight, go find your second number. Ask "how many of our Copilot seats have real activity," and if the answer is a shrug, that shrug is costing you every month. The tool that surfaces it matters less than the habit of refusing to renew per-seat AI licences on the purchased number alone.
The takeaway
Every Copilot rollout has two numbers, seats bought and seats used, and the whole renewal decision hinges on the second one, which is exactly the one most teams cannot see. Treat unused AI seats like idle infrastructure: provisioned, billing, doing nothing, and worth reclaiming. Before your next renewal, get the bought-vs-assigned-vs-used picture, look at who actually uses it, and put a monthly dollar figure on the seats that do not. Renew on evidence, not on the invoice.
If you have Copilot, do you know your second number, seats with real activity, or just the number you bought? And if you found the gap, how big was it? I keep hearing "bigger than we expected," which is the whole reason this is worth an afternoon.
Top comments (0)