Summary
Finance asked why the month cost 40 percent more when the seat count didn't change. You can now answer with evidence, and put controls in place so the next increase is a decision rather than a surprise.
What you learned
- Microsoft Copilot pairs a fixed per-seat license with usage-based billing. Only the metered half responds to tuning, and it moves independently of seat count.
- Every charge traces back through a billing account, a billing profile where the agreement uses one, and an invoice. Metered Copilot consumption follows Azure subscription and resource group scoping instead.
- Four inputs set what a single task costs: the model, the context, the tools, and the runtime. A forecast built from request counts alone misses.
- A pay-as-you-go billing policy defines funding and scope, but stays inactive until you connect it to a service. Scoping it to a specific group rather than all users is what keeps a pilot contained.
- Metered Copilot spend and Microsoft Foundry agent consumption sit on separate surfaces and don't combine into a single number. Reconciling a full picture means opening both and knowing which one reports credits rather than currency.
- The largest savings come before consumption happens. Decline work that doesn't belong to AI, route deterministic steps to rule-based logic, and match the model tier to the task.
- Concentrated consumption is a pattern to investigate, not automatically a problem to suppress. Heavy users who produce proportionate value are a template worth spreading, not a cost to cap.
Learn more
- Understand usage-based billing and cost management for Copilot Credits
- Managing AI experiences enabled by usage-based billing
- Usage-based billing: what you see in the Microsoft 365 admin center vs. your Azure bill
- Purchase and manage Copilot Credits
- Copilot Studio billing rates and management
- Select a primary AI model for your agent
- Plan and manage costs for Microsoft Foundry