M365 Copilot costs: included vs metered, and how to charge it back
The Copilot invoice now has two layers, and most cost models were built for one of them. This guide covers what each layer contains, why a seat count alone cannot allocate the metered line, and the one-page record that settles the allocation argument and names exactly when it reopens.
The bill has two layers, and the framing matters
Start with the sentence that gets mangled in most organisations: Copilot did not switch to metered billing. The flat per-seat licence, $30 per user per month on Microsoft's list price, did not change. What happened in June 2026 is that an additive metered layer arrived alongside it: Cowork credits and agent runs bill on consumption, at $0.01 per credit on Microsoft's list as of June 2026, on top of the seat price.
Those two framings lead to different meetings. "Copilot went metered" starts a renegotiation. "A metered layer arrived alongside the seats" starts an allocation exercise, which is the correct meeting, because the seat half of your invoice still allocates exactly the way it always did.
What is included, what meters
The seat covers Copilot in the apps: the assistant in Outlook, Word, Excel, PowerPoint and Teams that the licence has always described. The metered layer covers consumption-priced work: Cowork sessions and agent runs, billed in credits. Three facts about that layer decide most of the governance questions. It is off by default. An admin has to enable it. And spend controls ship with it, at tenant, group and user level, with usage alerts.
On the agent side, Microsoft publishes one worked example worth knowing: a Copilot Studio message runs one credit classically, two generatively, and a prompt grounded in the tenant graph carries a ten-credit surcharge. The unit price is public. What your tenant will consume is not, which is why every number in your cost model should come from your own invoice lines rather than from a projection.
Why a seat count alone cannot allocate the metered line
A credit pool is not a seat count, and "evenly" is not a method. Credits are driven by usage, not by who holds a licence, so allocating the metered line by headcount does not allocate it. It averages it, and the average moves money between business units without anybody deciding to.
A worked example, from an invented company with illustrative numbers and real arithmetic: an $80,000 monthly invoice splits into $54,000 of flat seats and $26,000 of credits across four business units. Split the credit half by seats and the heaviest consumer gets subsidised by roughly $2,474 a month, paid by units that barely touched the metered features. Nobody chose that transfer. The method did.
Showback, chargeback, and the gate between them
Showback reports each unit its share. Chargeback moves the money. The switch from one to the other deserves a materiality test: is the metered line big enough to be worth allocating at all? But clearing that gate is not permission. It tells you the line is worth allocating. It says nothing about whether every dollar inside it can be defended to the unit paying for it, which is what attribution coverage measures.
Two rules hold whichever method you land on. The shared-agent pool gets carved out and reallocated by seats, not by consumption, because shared infrastructure billed to its best-instrumented user is how that team ends up paying twice. And if your finance partner already has a materiality threshold, borrow it and cite it: borrowed thresholds survive audit better than invented ones.
The one-page record finance signs, and when it reopens
Four months from now, nobody will remember why the split works the way it does, and the unit carrying the biggest number will ask again. What survives is one signed page:
- 01The gate result, as a percentage, with the date it was measured.
- 02The verdict: showback only, or chargeback.
- 03The method for each line type: the per-seat layer, attributed credits, and the shared-agent pool.
- 04The conditions that reopen the decision. The two worth naming: the metered line crossing your threshold in either direction, and attribution coverage crossing your bar in either direction.
- 05Who decided, the date, and the next review date.
A signed page like that is the difference between a chargeback model and an opinion. The answer to "why does my unit carry this?" is written down, dated, and signed by whoever owns the budget.
The built version of everything above
The Copilot Chargeback Playbook is this guide, made runnable: a 12-question checklist that lands on a defensible allocation method in about 20 minutes, a 10-tab workbook you populate with your real invoice lines until the reconciliation tab ties to the invoice within a dollar, and the one-page policy finance signs. A working chargeback model in one afternoon. $49, one-time.
Get the Chargeback PlaybookOnly need to know whether you are ready to allocate at all? The free chargeback readiness checklist is the 10-minute version.
Prices and mechanics reflect Microsoft's published list as of June 2026 and can change; read your own invoice and admin center before you build. Kesslernity is an independent publisher, not affiliated with Microsoft. Practitioner guidance, not financial advice.