Microsoft just put a meter on Copilot. Who in your business is reading it?

4,000 is a strange number to find buried in a licence setting.

It is the number of Copilot Credits each user can burn through per month, by default, on any new Microsoft 365 Copilot Business licence bought through a reseller from 1 December. At Microsoft's list price of a cent a credit, that is up to $40 a user, every month, on top of the seat. Nobody has to approve it. Somebody has to switch it off.

For a 40-person business, the ceiling is about £1,200 a month. That's more than the Copilot seats themselves cost. And it arrives on a partner invoice that most finance teams pay without reading past the total.

I don't think most firms will hit that ceiling. I do think most won't know where they are against it, and that's the problem worth fixing, because Copilot is just the most visible example of something that's been happening across the whole software bill for two years.

On 1 October, Microsoft posted an update in Partner Center, the portal its resellers use [1]. The key lines:

Microsoft's own Copilot Credits licensing page sets pay-as-you-go at $0.01 per credit, billed monthly in arrears [2]. At the Bank of England's 8 October rate, $40 is about £30 [3].

Two things it doesn't do. It doesn't change existing licences automatically: this is about new purchases. And it isn't a price rise on the seat. It's a meter bolted next to it.

It lands while adoption is still climbing. The ONS's latest business survey, published 8 October, puts AI use at 30% of UK businesses in September 2026, up from 9% three years earlier. Among AI users with ten or more staff, 38% said admin and clerical roles were the most affected, and only 6% said AI had reduced headcount [4]. So firms are spending on AI, mostly to take load off the back office, and very few can yet point to a cost that went away. That's exactly when a usage meter needs someone watching it.

Microsoft's change is new, so the loudest reaction so far comes from the partners who'll be sending the invoices.

"Set your own limit now." Universal.cloud, a CSP partner, wrote up the change on 17 September, when the date was still 2 November (a partner blog, so it has an interest in selling you the management of this) [5]. Their warnings are worth repeating: Cowork is available to all users by default and users can install it themselves; "auto-apply new services" is on by default, so future agents can draw on the same budget; and without alerts, the first anyone hears of it is the monthly invoice. Their suggestion that finance gets a read-only role to watch consumption is the most sensible line in the piece.

"Conflicting figures." A WindowsForum news write-up (not a discussion thread) points out that earlier partner coverage quoted a $10-per-user limit, so anyone who budgeted on that number is now out by a factor of four [6].

On the wider software bill, owners have been grumbling for a while.

"What are we paying for that we didn't have in 2022?" That's a UK charity on the QuickBooks community in December 2025, facing a jump from £32 to £56 a month plus VAT from January. The official reply listed "improvements". The next reply, from a long-standing member, went through them one by one and was not persuaded [7].

"All of it eats hours." In a UK Business Forums thread from February and March 2026 about rethinking software, one owner listed the jobs that still swallow time: quoting, chasing invoices, onboarding, "moving data between systems, and reporting" [8]. That's worth holding onto, because it's the list of jobs AI is being sold to fix.

Meanwhile Xero's UK prices went up on 1 September: Ignite from £16 to £18, Comprehensive from £50 to £55 a month [9]. None of this is outrageous on its own. Together, it's a bill that moves every quarter with nobody in charge of it.

Take a professional services firm with 40 staff on Copilot Business. Compare the Cloud put the UK list price at £16.10 a user a month in February 2026 [10], so the seats cost about £644 a month.

Now the meter:

The usage mix is illustrative, not measured. The price and cap are Microsoft's [1][2]; the conversion uses the Bank of England's rate of 0.7561 for 8 October [3].

The plausible month adds about 30% to the Copilot bill. Fine, if it buys something. The ceiling nearly triples it, roughly £14,500 a year of exposure that nobody signed for. The true number for your business sits somewhere between those two lines, and right now it's on a CSP invoice a month after the event.

The real point is that you can't judge either number without a denominator. £197 a month is cheap if it saves your credit controller a day a week. It's expensive if it's two people asking Cowork to tidy their inbox.

This is a two- or three-day build in Power BI or Excel, using data you already have.

1. Sources

2. Model

A small star schema:

The unglamorous part is a vendor mapping table: rules that turn "MSFT*E0400ABCD" and "Microsoft Ireland Operations" into one vendor. Expect 30 to 60 rules, and two hours.

3. Measures

4. Alerts (Power BI data alerts or Power Automate, emailed to whoever owns the budget)

5. Dashboard

Three pages. The bill: monthly software spend stacked by seat and metered, with a 12-month trend. The vendors: a table with owner, utilisation, cost per active user and next renewal, sorted by cost. The AI page: credits by group against limits, projected month end, and cost per hour saved.

Caveats. Reseller invoices arrive in arrears, so the tracker sees last month, not today: the admin-centre consumption view is your early warning. Microsoft 365 usage reports can hide user names by default (a privacy setting), so work at group level unless you've agreed otherwise. Keep everything net of VAT, or the variance alerts will chase their own tail.

Capping at zero. It's tempting to set the limit to nothing and move on. Then the person who'd have got real value from Cowork stops trying, and you've paid for a seat to keep a chatbot on a leash.

Per-user caps that punish your best people. The heaviest users are often the ones getting the most out of it. Cap by group and review the outliers, rather than throttling them by default.

A tracker that becomes another subscription. If nobody owns the alerts, you've built a very tidy record of money you didn't save. Name one person.

Mistaking a meter for a scandal. Usage pricing isn't a con. It's fairer than paying for seats nobody opens. The risk isn't the model; it's nobody reading the meter.

If you'd rather have the tracker built on your own ledger and admin exports, with your vendor mapping already done, that's a one-week Data Clarity Sprint. Reply and I'll send the outline.

— Aksel