What changed
GitHub added a new "Potential return on investment" section to the Copilot impact dashboard on August 7, 2026. The new view compares lighter Copilot users against more agent-first cohorts and shows three concrete fields side by side: estimated cost per developer per month, that cost as a share of payroll, and pull requests per developer per month. GitHub also added a salary selector so teams can model the numbers against their own compensation assumptions instead of treating the dashboard as a fixed benchmark. The release note is here: GitHub changelog.
Why it matters
This is a product shift from "is AI being used?" to "is deeper AI usage paying off?" GitHub's own docs say the impact dashboard groups users into adoption cohorts based on how they engage with Copilot and connects that adoption to pull request throughput, rather than relying on flat active-user counts alone. Its usage-metrics docs go further and define those cohorts around progressively deeper agent usage, including CLI, cloud agent, and the Copilot app. Together, that means GitHub is treating AI adoption depth, workflow output, and cost as one management surface instead of three separate reports. See the supporting docs for the Copilot impact dashboard and Copilot usage metrics.
The builder takeaway
If you build AI features, the important lesson is not the specific GitHub dashboard. It is the measurement model behind it. Usage-based AI products are maturing past seat counts and chat volume. The next product layer is cohorting users by depth of workflow change, then tying spend to delivery output that a team already understands. GitHub is also keeping a raw-data escape hatch through its metrics APIs and NDJSON exports, including team-level reporting guidance for custom BI workflows, which is a good pattern for any AI product that wants to be trusted by engineering leaders. Reference: team-level Copilot usage metrics.