Chapter 1 of 6
Module scope#
What we will do#
Objective. Make AI spend attributable before finance asks for the monthly split.
Deploy cost-allocation tag policies, a filtered AI budget, and one API Management token chargeback policy for the approved AI gateway route. Record the token rates agreed with finance. The check then runs a KQL query that groups tokens by cost center, consumer, product, API, and model.
Why it matters#
Problem. Azure Cost Management shows billed cost, but a shared AI gateway can hide which product or cost center drove token use.
Solution. Tags bind billed cost to an owner. The gateway emits token metrics with finance-owned dimensions. The rate table gives FinOps a clear way to turn attributed token use into showback or chargeback.
EU AI Act. No article requires cost allocation; this module is good practice. For high-risk AI systems, the per-consumer token records can support Article 12 (record-keeping). This is an engineering mapping, not legal advice.
Boundaries#
This module sits outside the numbered session sequence.
It changes the approved AI FinOps subscription, the filtered AI budget, and the existing API Management AI gateway policy source. Azure Policy remains authoritative for policy assignments. Cost Management remains authoritative for billed cost. Application Insights or Log Analytics remains authoritative for token telemetry and chargeback queries.
Session 06 owns the gateway route. Session 11 owns the wider observability and cost notification baseline. This module adds finance-grade allocation on top of those controls. Fabric capacity, Copilot Credits, FinOps hubs, PTU reservations, and reservation chargeback are outside this module.
Module preparation
Who should join
- FinOps and cloud platform engineers
- Azure API Management gateway owners
- Application Insights and Log Analytics operators
- AI product owners and finance partners
What you need
- Session 06 has an approved Azure API Management route for the AI API, and the gateway owner controls the versioned policy source.
- Session 11 has the approved Application Insights or Log Analytics telemetry boundary, action group, and cost-notification owner.
- The target Azure subscription is recorded as a full Azure subscription resource ID, and the cost owner approves the filtered budget scope.
- The required cost-allocation tag names, cost-center taxonomy, APIM products, consumer mapping, and monthly token quota are agreed with finance.
- API Management is integrated with Application Insights, custom metrics with dimensions are enabled, and the LLM API diagnostic settings can support the chargeback query.