3.4. Monitor AI Services in Microsoft 365
💡 First Principle: AI services are metered, adopted unevenly and occasionally unavailable — so an administrator monitors three separate things: cost (what's being consumed and who pays), usage (who is actually getting value) and health (whether the service is working). Each has its own reports, and mixing them up gives wrong answers.
Why care? Licenses are expensive and consumption billing is open-ended. A tenant can pay for thousands of Copilot seats nobody uses, or run up usage-based charges no one noticed. Leadership will also ask whether Copilot is working before renewing it. Monitoring is how you answer with evidence instead of anecdote. Think of it as three gauges on one dashboard: the bill (cost), the passengers actually on board (usage) and the engine warning light (health).
Microsoft frames this within Copilot controls (formerly the Copilot Control System), whose three pillars are security and governance, management controls (licensing and metering, agent lifecycle, customization) and measurement and reporting (readiness, adoption, impact and value — mostly brought together in Copilot Analytics).
⚠️ Common Misconception: "Enabled users" in the Copilot usage report doesn't measure adoption — it counts licensed users. Adoption is active users ÷ enabled users.