BillingAzure FOCUS export
Billed Azure OpenAI and Foundry model cost, in the same cost model as the rest of Azure.
Boundary Does not infer model, token, deployment, or business outcome detail that the export does not contain.
Microsoft AI cost evidence
Separate the Microsoft AI bill from its usage signals.
CloudMonitor starts with billed Azure OpenAI and Foundry cost. Deployment and Copilot usage stay separate, appearing only through a documented CloudMonitor connection.
The evidence ledger
Follow each Microsoft AI claim back to its record.
The bill establishes cost. Operational and license feeds add detail, but only for the scope and period they cover.
OperationsAzure Monitor reference
Microsoft exposes deployment, token, PTU utilization, and spillover metrics through Azure Monitor.
Boundary This page does not claim a production CloudMonitor connector. Verify current connections before including these metrics.
Developer toolsGitHub Copilot connection
Assigned seats, activity, adoption, and supported usage detail after the GitHub connection syncs.
Boundary GitHub remains authoritative for rates, licenses, and the invoice.
ProductivityMicrosoft 365 Copilot
The Microsoft charge can identify Copilot cost when the billed product is present.
Boundary Licensed-user activity is a separate Microsoft Graph reporting source; it is not inferred from the charge.
Check where your numbers come from before using a signal in an allocation or efficiency calculation.
A PTU decision, not a dashboard verdict
Price steady capacity against the traffic you must serve.
Provisioned throughput is billed while it is deployed. A sound decision also needs observed demand, service requirements, and the standard token cost of any configured spillover.
Steady capacity Deployed PTUs × hourly meter
Peak traffic Standard spillover token cost
Service guardrails Latency, throughput, availability, headroom
- 01Observe
Use a representative demand window. Separate normal load, peaks, and expected growth.
- 02Price
Compare the active PTU meter with standard usage and any spillover at the same service scope.
- 03Protect
Keep deliberate headroom where latency, throughput, or availability requirements justify it.
- 04Approve
Let the workload owner accept the capacity and service tradeoff before a change is made.
Keep the decision boundary visible
CloudMonitor organizes the evidence. Your owners decide.
CloudMonitor is advisory and read-only. It does not resize deployments, change licenses, or turn an operational signal into approved business value.
- CloudMonitor
- Reads connected FOCUS cost, applies cost groups, and reports connected GitHub Copilot usage. CloudMonitor runs on Microsoft Fabric: in your own Fabric tenant by default, or CloudMonitor-hosted. Architecture and data flow sets out where each layer runs in both arrangements.
- Workload owner
- Confirms the model, deployment, service target, capacity, spillover, and license decisions for the workload.
- Finance and product
- Reconcile currency and invoice scope, then supply the approved investment, outcome, and comparison baseline.
Source: this page interprets the FinOps for AI technology category published by the FinOps Foundation, licensed under CC BY 4.0. The wording, examples, and product mapping on this page are CloudMonitor’s own.
Build the Microsoft AI baseline your evidence supports.
Connect Azure billing first. Add operational and Copilot usage only where a supported source is available.
