On October 19, 2026, every new Microsoft 365 Copilot Business subscription starts arriving with usage-based billing already switched on. Microsoft announced the change in message center post MC1476200, “Simplifying AI access: introducing usage-based billing,” and the Cloud Solution Provider channel follows on November 2 (EPC Group’s breakdown of the rollout). For new seats, Microsoft automatically creates the Azure resources, attaches a default spending policy, and leaves the “Auto-apply new services” toggle on, so future agents start drawing down credits the day they ship.
Existing subscriptions do not change. But any Copilot Business seat you buy after that date is metered out of the box, and the default controls are Microsoft’s, not yours. That distinction is where the next surprise bill lives.
What actually flips on October 19
Three things happen automatically on new eligible purchases. Microsoft provisions an Azure subscription to carry the charges, applies a tenant-level spending policy with its own default ceiling, and turns on auto-apply so that any newly supported Copilot agent or service is swept into that policy without a second decision from you. The Microsoft 365 admin documentation is explicit that “Auto-apply new services” ships in the on position for both the default policy and any new policy you create (Microsoft Learn).
The timing is not an accident. Fewer than 7% of Microsoft’s roughly 450 million commercial Office seats had adopted Copilot when the company described the shift, which is the gap a consumption meter is designed to close (SaaSrise). It lands on top of a list-price reset that already raised Microsoft 365 commercial suites 5 to 43% on July 1, folding baseline Copilot Chat into most tiers so generative AI is a built-in cost rather than an optional line (SaaSrise). Your per-seat cost went up, and now a meter runs behind it.
Two meters, two very different bills
Copilot’s usage charges split into two systems, and they behave differently. Finance teams that treat them as one line will mis-model both.
The first is Copilot Credits, priced around $0.01 per credit, which cover Copilot Cowork, Cowork-built apps, and the Work IQ API. These run through the Cost Management dashboard in the Microsoft 365 admin center. Cowork, the long-running autonomous agent, is disabled by default, which is the one piece of good news in the setup.
The second is the Copilot Studio message meter, billed at $0.01 per message, and it is the one that quietly adds up. It covers Copilot Chat, SharePoint agents, and any custom agent or Copilot built in Copilot Studio (Microsoft Learn meter table). Microsoft defines a billable message as any request or response that triggers an action, so a single conversational agent handling routine questions can log dozens of billable messages per user per day. Those charges do not sit in a tidy SaaS invoice. They land on the linked Azure subscription.
The budget gap finance keeps missing
Here is the part that matters most, and the part the press coverage skips. The two meters do not enforce limits the same way.
Copilot Credits policies can hard-stop. When you set a limited monthly budget on a credit policy, Microsoft’s own documentation states that “when users hit the limit, they lose access to agents and services for the rest of the month until credits reset on the first of the month” (Microsoft Learn). That is a real cap.
The Copilot Studio meter does not work that way, because it runs through an Azure subscription, and Azure budgets alert without enforcing. Reaching 100% of a budget does not stop the service or the billing. The only way to stop the charges is to disconnect the service (EPC Group). So the services most likely to generate runaway spend, the chat and custom agents people actually use all day, are governed by a budget that only watches the number climb.
This is the same visibility problem that already lets a third of AI spend sit outside finance’s view, except now the meter is bolted onto a product you already pay a seat for. A budget alert is not a control. It is a notification that the control you assumed you had does not exist.
Set these before the 19th
If your tenant buys Copilot Business seats in the next quarter, treat the default policy as a draft you inherited, not a decision you made. A short pass through the Cost Management node closes the biggest gaps.
- Replace Microsoft’s default ceiling with your own monthly limit, and set a per-user cap so one account cannot drain the pool. The admin center recommends the per-user limit for exactly this reason.
- Decide “Auto-apply new services” on purpose. Turn it off if you are in a regulated environment or want to vet each new agent before it can bill.
- Put the Copilot Studio meter on its own Azure subscription or resource group, give it a dedicated budget, and route alerts to both IT and finance at 50, 80, and 100%. Document the disconnect procedure, because disconnecting is the only hard stop that exists on that path.
- If you carry a Microsoft Azure Consumption Commitment, point the billing subscription at the account that holds it so eligible Copilot usage draws down the commitment instead of adding net-new spend (Microsoft Learn).
- Fold the consumption into your existing chargeback or showback model from day one, so the spend has an owner before the first invoice.
The seat-era control meets a consumption-era bill
The State of FinOps 2026 survey found 98% of FinOps teams now manage some form of AI spend, up from 63% a year earlier and 31% two years before that, across 1,192 practitioners representing more than $83 billion in annual cloud spend (economize). A default-on meter with an alert-only budget is precisely how that spend grows without anyone choosing to grow it.
None of this is an argument to turn Copilot off. It is an argument to recognize that pricing moved. The per-seat license you negotiated is now a floor, and the real variable cost rides a consumption meter inside it. That is the same shift HubSpot made when it started billing per resolution, and the same reason your dashboard’s token count is not the number your CFO wants. Seat-era governance caps headcount. It does not cap usage. On October 19, the gap between those two becomes your problem to close, and the clock on the default policy is already running.