GitHub adds client-level Copilot usage metrics—but request counts are not developer productivity

The new fields describe traffic, sessions and model routing

Each client entry can include successful model-request counts, distinct sessions, prompt and output tokens, average tokens per request and a breakdown by model. A `model_selection_mode` identifies a user-selected model, automatic routing or a product-selected model; other or unnormalized values may appear as `other` or `unknown`.

The reports require an enabled Copilot usage metrics policy and appropriate access, such as an enterprise owner, billing manager, organization owner or a custom role with the View Copilot Metrics permission. In 28-day aggregate reports, GitHub says the client totals appear inside each day’s totals rather than as one undated lifetime number.

Related sources: GitHub: usage metrics API adds client reporting ↗

One prompt can create several model requests

GitHub explicitly says `request_count` measures model calls, not prompts. An agent may make several calls while working on one request, and a short interactive completion may produce a very different token pattern from a longer coding task. Client identity is also a best-effort grouping supplied through an integration ID, not a verified statement about the person or business purpose behind the work.

That makes direct productivity rankings unsafe. A team with more requests may be doing more useful work, retrying failures, using a more agentic client or simply routing the same task differently. Token volume is useful for capacity and cost analysis, but it does not establish code quality, time saved or adoption satisfaction.

Related sources: GitHub: usage metrics API adds client reporting ↗

Canadian teams should connect usage to reviewed outcomes

For Canadian organizations, the useful governance view combines traffic with repository policy, approved-client lists, task outcomes and human review. Segment by client and model selection, then compare accepted pull requests, review findings, rollback rates and support incidents without turning individual telemetry into an employee score.

Document who can see the reports, how long exported data is retained and how metrics will be interpreted. A newly visible client may identify a licensing or support gap, but it should trigger investigation rather than an automatic compliance conclusion. The API is an operational measurement tool, not an independent study of Copilot’s business value.

Sources and editorial method

We use the linked original statements and reports to distinguish documented facts from our analysis. Company announcements describe their authors’ plans and claims; they are not independent verification of future outcomes. This article is dated and will be revised if material evidence changes. Read our editorial standards and corrections process.

1. GitHub: usage metrics API adds client reporting ↗ — October 23 announcement; fields, scopes, access requirements and interpretation limits.

Continue reading

Explore the Canadian AI guide, read the dated source monitor, or run an internet speed test to establish a baseline for your own connection. That test cannot measure any AI provider’s model quality.