Reconcile Copilot and agent costs across surfaces
Usage is measured now. But leadership's other question was cost—where do you even look for that, and is it the same place you just looked for usage? Before you open anything, predict the answer: does one single report show every dollar Relecloud spends on Copilot and its Woodgrove fraud-investigation agent together? It doesn't. Costs split across surfaces by what incurred them, not by which business problem they support—and that split is the whole reason a monthly readout takes more than one dashboard to build.
Two cost views inside the Microsoft 365 admin center
The Microsoft 365 admin center holds two related but distinct places to look at Copilot spend, and they answer different questions. Billing > Cost Management gives you a service-level view of costs based on Copilot usage—a summary of what Relecloud's pay-as-you-go Copilot services cost in total. Reaching it requires the Global Administrator or Billing Administrator role.
Copilot > Billing & usage > Billing policies gives you something narrower and more operational: per-policy budget and spending detail for a specific Copilot pay-as-you-go service. You don't need Global Administrator to work here day to day. The AI Administrator or License Administrator role lets you create spending policies, manage limits and alerts, and view the billing-policy dashboard. Only Global Administrator or Billing Administrator can set or change the billing method itself—connecting or disconnecting the Azure subscription behind a policy. That distinction matters for least privilege: most of the people who monitor Relecloud's Copilot spend month to month never need the keys to the billing method, only the ability to watch the dashboard and manage a budget.
A third surface tracks the same spend at the credit level rather than in dollars: the Copilot Credits usage report, under Reports > Usage in the Microsoft 365 admin center, shows metered Copilot Chat credit consumption against your pay-as-you-go billing policies. Reach for it when you want to see credits used rather than the dollar totals the Cost Management and Billing policy views report.
| Cost surface | What it shows | Role |
|---|---|---|
| Billing > Cost Management | Service-level total of pay-as-you-go Copilot costs | Global Admin or Billing Admin |
| Copilot > Billing & usage > Billing policies | Per-policy budget and spending for a specific Copilot service | AI Admin or License Admin (Global/Billing Admin to change the billing method) |
| Copilot Credits usage report (Reports > Usage) | Metered Copilot Chat credit consumption | Admin center reports access |
| Microsoft Cost Management | Microsoft Foundry agent consumption (tokens, tools, infrastructure) | Azure resource-group access |
With those two views in mind, the next question is what actually feeds the billing-policy side of that picture.
How a billing policy connects spend to a service
A billing policy is the mechanism that links an Azure subscription and resource group to a specific Copilot pay-as-you-go service—Microsoft 365 Copilot Chat or SharePoint agents, for example—for a defined group of users. Setting one up is a two-step process, and stopping after step one leaves billing incomplete.
- Create the billing policy. Name it, choose or create an Azure subscription and resource group to associate with it, pick a region, and accept the terms.
- Connect the policy to a Copilot service. The policy itself only defines where charges land—it doesn't start metering anything until you connect it to the service you want to bill through it.
That two-step design is why a policy that exists but isn't connected shows no useful spending data: the link between "who pays" and "what's being used" doesn't exist yet. Once connected, the policy's Budget and Spending views start reflecting actual Copilot usage costs for that service.
None of this, though, touches the other half of Relecloud's monthly readout—the Woodgrove fraud-investigation agent.
Why the Woodgrove agent lives in Microsoft Cost Management instead
Microsoft Foundry agents, including Woodgrove's fraud-investigation agent, are billed as ordinary Azure resource-group consumption, not through the Microsoft 365 admin center's Copilot billing policy at all. There's no flat per-agent fee. Charges reflect what the agent actually consumes in the background: model usage (tokens spent on inference), tool usage (search calls, functions, connectors), and infrastructure usage that supports the agent's runtime.
That's the resolution to the predict-before-reveal question from the top of this unit. The Woodgrove agent's cost never appears in the Copilot billing-policy dashboard, because it was never a Copilot pay-as-you-go charge to begin with. You find it in Microsoft Cost Management, scoped to the resource group or the specific Foundry resource, using Cost Analysis and filtering or grouping by resource or by meter. Reconciling Relecloud's full AI spend means opening both surfaces and adding them together yourself—no dashboard combines them for you, because they're tracking two fundamentally different kinds of billing.
Guiding question: If a colleague tells you the "total Copilot and agent cost" is sitting in one place, what would you ask them to check before trusting that number?
Budgets and alerts: the safeguard between reconciliations
You don't reconcile costs only once a month. A billing policy's Budget tab lets you set an ongoing safeguard between manual checks. Configure a dollar spending limit, a reset cadence, and email alert thresholds:
- Monthly resets the limit on the first of each month.
- Quarterly resets on January 1, April 1, July 1, and October 1.
- Yearly resets on January 1.
Alert recipients must be mail-enabled security groups, not individual users, which keeps the recipient list manageable as people join or leave the monitoring rotation. A default threshold fires at 100% of budget, and you can add up to four more thresholds anywhere between 1% and 99%—early warnings before the limit hits.
Here's the rule worth remembering above every other detail in this unit: reaching 100% of budget doesn't stop the service or the billing. It sends a notification, nothing more. Copilot usage—and Woodgrove's agent consumption in Azure—keeps running and keeps accruing cost past that threshold unless someone acts on the alert. The only way to actually stop billing is to disconnect the pay-as-you-go service from the policy. A budget alert tells you the door is open; it doesn't close it.
Two timing details matter when you're watching these numbers closely. Spending data in the admin center's budget Spending view can take up to four hours to appear, and alert emails themselves can lag up to 24 hours behind the spend that triggered them. If Relecloud's finance team asks why an alert landed a day after the spike, that latency is the answer, not a missed configuration.
Reflect: Suppose Relecloud's billing policy hits its 100% alert threshold overnight, and the Woodgrove agent keeps investigating fraud cases the next morning at the same pace. Why doesn't the alert itself stop that from happening, and what's the one action an administrator takes if they need a hard stop instead of a notification?
Usage is measured, cost is reconciled. Is there one place that pulls all of this—security, management, and reporting—together at a glance?