Microsoft Says 30 Million Paid Copilot Seats Is the Wrong Metric to Watch

Thirty million paid Microsoft 365 Copilot seats is the headline number from Microsoft’s latest earnings call, but the company’s own follow-up post argues seat count isn’t the metric that matters anymore — work transformed is. Net seat adds more than doubled quarter over quarter, and Microsoft says the real signal is usage depth: conversations per user nearly doubled year over year, users touching multiple Copilot features are up triple digits, and weekly engagement now matches Outlook and Teams. Customers with more than 50,000 seats grew over 7x year over year, and the time to reach 80%+ monthly active usage has compressed from months to days. The examples Microsoft cites are concrete: its own Cloud Supply Chain team cut cycle time 75% using more than 70 agents, Eaton accelerated manufacturing root-cause analysis across roughly 5,000 reports, and EY’s Autonomous Sourcing Agent has handled 200+ procurement transactions since an October 2025 pilot, on pace for 1,500 within a year.

That last data point — an agent embedded in one workflow, expanding its share of the work over time rather than replacing a role outright — matches how Microsoft is shipping the rest of its Copilot stack. Service Agent, the Dynamics 365 Customer Service agent Microsoft took to general availability in mid-July, ships with more than 90 new MCP tools spanning case management, customer insights, and quality coaching, and is explicitly built to act across Outlook and Teams while retaining case context — the same cross-app reasoning pattern Microsoft is now pointing to as evidence of “work transformed.”

The pattern worth watching isn’t adoption, which Microsoft has clearly cleared. It’s whether the agents that show up inside existing tools — sourcing, service, supply chain — keep expanding their share of a workflow the way EY’s has, or plateau once the easy transactions are gone. Seat counts answer “did companies buy it.” Transaction-volume trajectories like EY’s are what actually answer whether the AI is doing more work over time.