Full narration of the briefing. Click a section heading to jump the player to that moment.
Everything in your Microsoft agreement so far has been per user. You count people, you multiply, you know the bill. The layer arriving beside Copilot does not work that way. Cowork bills per task.
Agents consume credits. And credits are bought by the block or drawn down as you go. I am Daniel, Claire has the numbers, and this briefing is about the part of your 2027 renewal you cannot forecast from a headcount, which is exactly why it needs terms around it before you sign rather than after the first surprising invoice.
Let me lay the stack out, because the pieces are easy to confuse. At the bottom, your Microsoft 365 suite, thirty nine dollars on E3 or sixty on E5 after the July rise. Above it, the Microsoft 365 Copilot seat at thirty dollars, which is what gives a person Copilot inside Word, Excel, Outlook and Teams. Above that, Copilot Cowork, which launched globally and bills per task on top of that thirty dollar seat.
And underneath all of it, Copilot Credits, which is how Copilot Studio agents, Agent 365 and Cowork actually meter, at roughly one cent per credit on pay as you go, and cheaper if you pre commit to a block. Four layers, and only the first two are predictable from headcount.
Agent 365 deserves a moment on its own, because it answers a question most enterprises have not asked yet. Once agents are doing work in your tenant, they need what employees need: an identity, permissions, an audit trail, a compliance boundary and someone accountable for them. Agent 365 is that administration layer, per seat, and it is included inside E7. Whether you need it in 2027 depends entirely on whether you are actually running agents at scale by then.
If you are, it is not optional. If you are not, it is a solution to a problem that has not arrived, and paying for it early is the most common way this whole category gets oversold.
Here is what makes this genuinely difficult, and it is worth saying plainly. With seats, cost scales with headcount, which you know. With credits, cost scales with usage, which you do not, and which grows precisely when the technology works. A successful agent gets used more.
So the better your deployment goes, the faster the meter runs, and the business case that justified it can be undermined by its own success unless you modelled the upside. Nobody can hand you an accurate two year credit forecast in 2027, including Microsoft. What you can do is refuse to sign a number that assumes anybody could.
So estimate from measurement rather than from a spreadsheet. Take one agent, in one process, and run it for a month with the meter visible. Record credits consumed per completed task, and tasks per month. Now you have a unit rate you own, and you can extrapolate honestly to the processes you plan to automate next.
That is a real forecast with a real basis, and it will be more conservative than the one in the proposal. Note also what the meter reaches beyond credits: the underlying data, storage and connectors your agents touch have their own costs, and those land on a different line of your bill.
Then negotiate four things, because on a meter the terms matter more than the rate. A rate hold, so your credit price is fixed for the whole term and growth does not reprice you. Rollover, so credits you bought and did not burn carry into the next period instead of expiring as pure margin. An overage price agreed and capped in advance, because unpriced overage is where consumption contracts go badly wrong.
And spend alerting with hard thresholds, so finance learns about a spike in the week it happens rather than in the quarterly review. Buy credits in tranches against proven consumption, never in one large block sized on optimism.
The move from this briefing: before your renewal, put one agent into one real process and measure credits per task for a month. That single number turns every consumption conversation from a debate about projections into arithmetic you own. And if the answer is that you are not ready to run agents at scale in 2027, that is a perfectly good finding, and it saves you from committing to a meter you will not use. Next session: Azure and the MACC, which is where the leverage in this whole negotiation actually sits.
See you there.
Redress Compliance works on contingency: our fee is 25 percent of what we save you. Nothing saved, nothing paid. Independent, buyer side only, never vendor funded. Want Redress to contact you? Reach out and we respond the same day.
Talk to a Microsoft negotiator