AI, in your work.

What the announcement adds

On October 22, 2026, GitHub announced the Industry comparison section in the Copilot impact dashboard. It compares an organization's own indicators with aggregated benchmarks from other participants. Participation is optional and disabled by default, according to GitHub's announcement.

The access guide covers enterprises and organizations that are not owned by an enterprise. Enterprise-owned organizations participate through their enterprise; enterprises used solely to assign Copilot Business licenses are excluded. The Copilot usage metrics policy must be enabled.

Enterprise owners and owners of independent organizations can opt in from the dashboard or through the Copilot metrics industry comparison policy: both control the same setting. Results are visible to people who already have dashboard access.

Two indicators answering different questions

The comparison documentation distinguishes these indicators for the 28-day reporting period shown in the dashboard.

The classification reference requires qualifying activity on at least two distinct days within the trailing 28-day window. Agent-first corresponds to one GitHub agent surface; multi-agent means two or more, or the Copilot app. Classification is recalculated daily. Using IDE agent mode alone does not place someone in the agent-first group.

Who each organization is compared with

The announcement offers company size and ten industries as selections. If too few industry peers participate, the dashboard expands the comparison to all industries within the same company size. The median may appear first; the top decile requires a larger group.

GitHub specifies that company size and industry are self-reported and unverified. With data residency, comparisons remain within the same region. The documentation also says participants cannot see the identities or number of participants in each cohort.

A missing ranking can mean insufficient participants or an own score that cannot be calculated. Without non-agentic users, for example, leverage lacks a baseline. It does not mean a score of zero.

What participation involves

According to GitHub, each participant contributes two summary scores, without user names, code, repository content, or pull request content. Aggregates are recalculated daily and deleted after 30 days. Comparisons are excluded from the API, NDJSON exports, and downloads.

Opting out immediately removes comparisons from the dashboard; data stops contributing from the next daily update. The participation guide explains this timing.

Interpretation: a reference for investigation

GitHub warns in its interpretation guide that these comparisons show associations. Team composition, project complexity, and adoption maturity can influence differences. The participating peer group can also change between reporting periods.

Our reading is to use relative position to ask questions: which tasks agents receive, how changes are divided, and where reviews stall. A benchmark drawn from voluntary participants does not establish representativeness of an entire industry. A better ranking can also coexist with an unchanged own score when the comparison group changes.

Synthetic example: the same ratio can hide different work

Consider two fictional teams. The following rates are illustrative assumptions per user per month, not results from GitHub or its customers.

• Fictional team A: agentic users, 12 PRs per user per month; non-agentic users, 6 PRs per user per month; ratio, 2×. Assumed context: small changes to a web application.
• Fictional team B: agentic users, 4 PRs per user per month; non-agentic users, 2 PRs per user per month; ratio, 2×. Assumed context: large migrations requiring specialist review.

Turning the comparison into a decision

Both teams in the example have the same ratio while delivering changes of different scope. That equality does not establish equal value or show that Copilot doubled productivity. Work allocation and review requirements are part of the explanation that would need investigation.

As Nexo's editorial recommendation, before turning a benchmark into a target:

• Record the reporting period and the cohort actually used, particularly when it expanded to all industries.
• Check differences against task mix, change size, incidents, and team experience.
• Choose a specific intervention, such as training or improved review, and assess its evolution alongside delivery outcomes and developer feedback.