Summary

  • 3 minutes

In this module, you learned that:

  • A spending policy enables a usage-based AI service for a defined group of people (identity-based) or a defined set of agents (resource-based) - and nothing is billed until one exists.
  • When a user falls under more than one policy, the system applies exactly one in full—selected by the highest per-user limit, then the largest overall limit, and then the most recently created policy.
  • Admins set a monthly policy limit, an optional per-user cap, and alert thresholds. Users see banners as they approach a limit, and an in-progress task continues and finishes even after the user crosses the limit.
  • Users can request more capacity in-product, admins approve requests in Cost Management, and a request policy can route requests to an existing workflow without removing the admin's final say.
  • Creating a spending policy requires Global Administrator or Billing Administrator, who also approve credit requests alongside AI Administrator; AI Administrator and License Administrator can edit existing policies and manage limits and alerts; reader roles like AI Reader provide read-only access.
  • A resource-based spending policy caps what a specific agent can spend; Agent 365 closes the gap that cap leaves open, with a unified agent registry, lifecycle actions like activate or block, and a Microsoft Entra Agent ID that ties every agent to an accountable owner.

Caldova now has a spending policy scoped to its pilot department, limits and alerts that keep it within budget, a way to handle requests for more capacity, admin roles assigned on a least-privilege basis, and a registry check to confirm every agent has an owner. You can apply these controls in your organization to manage usage-based AI consumption and spending.

Learn more