Quick answer
The useful Copilot stats are active seats, credits consumed per active seat, acceptance rate for suggestions, chat versus agent usage, and model mix. Suggestion counts and raw acceptance totals look impressive and say little about value. Measure output per seat, not activity per seat.
The metrics
| Metric | Measures | Watch for |
|---|---|---|
| Active seats | Adoption | Assigned but unused |
| Credits per active seat | Consumption | Model mix changes |
| Suggestion acceptance | Usefulness | Gaming by trivial edits |
| Chat turns | Exploration | Long loops |
| Agent runs | Automation | Repeated context |
| Model mix | Cost driver | Premium drift |
| Output per seat | Value | Hard to attribute |
Two misleading numbers
- Total suggestions: rises with typing, not with value.
- Total credits: rises with adoption even when efficiency improves.
Normalise both. Credits per merged change and suggestions per accepted edit are more honest.
Report the set that fits the audience
Engineering leaders want output per seat and cost per team. Finance wants spend and forecast. Developers want their own balance. One dashboard rarely serves all three, so publish a small set and link to detail.
UsageAtlas covers the developer view; the organization report covers the leadership view.
The four numbers worth watching
| Metric | Why it matters | Typical signal |
|---|---|---|
| Allowance used per cycle | Predicts blocking | Steady rise is healthy |
| Premium vs standard split | Explains cost | Shift means model change |
| Active seats vs paid seats | Finds waste | Idle seats |
| Requests per completed task | Measures efficiency | Rising means retries |
Averages hide the days that hurt. Look at the maximum day in a cycle as well as the mean, because one long agent session can consume a week of allowance.
The team analytics guide covers the organization version of these numbers.
Questions, answered
What is a good Copilot acceptance rate?
There is no universal target. A high rate on trivial edits is not better than a moderate rate on meaningful ones. Compare teams doing similar work.
Are Copilot stats available per repository?
Organization metrics are aggregated by editor and language, not by repository. Use case studies for repository-level detail.
Do Copilot stats include inline completions?
Yes, as suggestion and acceptance counts. They are usually not metered as premium usage.