GitHub said in an August 7, 2026 changelog entry that the Copilot impact dashboard now includes a “Potential return on investment” section 1. GitHub describes the section as connecting what you spend on Copilot to the pull request output you get back 1. Two comparison cards show metrics for different user adoption phases 1.
As spending on AI coding tools accumulates, so does the need to explain internally whether it is worth the money. What was added is an official set of numbers for that conversation. It is also built in a way that can mislead if the numbers are copied into an internal deck without their caveats, which GitHub attaches itself.
Three metrics
Three metrics appear on the comparison cards 1:
- Cost/dev/month — average monthly Copilot cost per developer
- % Payroll/month — cost expressed as a share of developer compensation
- Pull requests/month — average pull requests per developer per month
Two sit on the spending side, one on the output side. Spending is shown both as an amount and as a share of payroll, while the only thing on the return side is a count of pull requests. The changelog does not address why GitHub chose that metric, or whether pull request counts are a sound proxy for output.
The caveats GitHub attaches
The entry is explicit about how to read the numbers: cost figures are estimates based on AI credit consumption, and the salary selector is a modeling input rather than actual payroll data 1.
So “Cost/dev/month” is not the invoiced amount itself, and the denominator behind ”% Payroll/month” is not your organization’s real payroll data. Together with the section being named “Potential” return on investment, this reads as GitHub declining to present the figures as a settled ROI calculation. Whether those two caveats travel with the numbers changes what the numbers mean.
A counting change landed alongside it. GitHub improved adoption cohort accuracy by counting all users active during the full 28-day reporting window, rather than only those active on the final day 1. Because that shifts how headcount appears across the same period, comparisons against earlier figures need care.
Who can see it
Availability is at the enterprise and organization level 1. Access goes to enterprise owners, billing managers, organization owners, and those with custom roles granting the “View Copilot Metrics” permission, all requiring the Copilot usage metrics policy to be enabled 1. This is not a view individual developers use on their own numbers; it is a screen for the people managing the spend.
Part of an early-August run of Copilot management and measurement
The addition is not a one-off. It joins several management and measurement updates GitHub has shipped around Copilot in a matter of days.
On the same August 7, the Copilot usage metrics API gained a per-agent-app breakdown, making it possible to measure which agent was used how much — the counterpart to a dashboard that lines up spending against output.
On the depth-of-review dial, code review effort levels reached general availability on August 7. On governance, MCP server allowlists became enforceable through enterprise settings on August 6. Deciding what can be used, choosing how deep it goes, and measuring how it gets used all arrived in short order.
Before taking the numbers into a meeting
Having official spending figures gives comparisons a starting point. What the screen shows, though, is your own organization’s trend, not a cost-effectiveness comparison against other tools.
Three things are worth carrying alongside the numbers: cost is an estimate, the payroll denominator is a modeling input, and the output metric is a pull request count. Stated with those, a discussion that has largely run on impressions can at least run on shared figures.
Sources
- Copilot impact dashboard adds a return on investment section - GitHub Changelog (August 7, 2026)