What does the measure place side by side?
The section compares two groups of developers: those who remain in chat and code completion, and those working agent-first. Each card carries three things — the monthly Copilot cost per developer computed from actual AI credit consumption, that cost as a share of developer payroll, and the average monthly pull requests per developer.[1]
None of the three measures belongs to the tool alone. The cost belongs to the tool, the payroll to the employee, and the pull request count to how a team works. Gathered on one card, the resulting ratio reads as the tool's return, while what sits in the denominator is a person's salary.[1]
Which decision does the number make easier?
Administrators can enter their own payroll assumptions through a salary selector, and the cost-derived measures recalculate immediately. The announcement adds two caveats: the cost figures are estimates based on credit consumption, and the salary selector should be treated as a modelling input rather than actual payroll data. The section appears at enterprise and organization levels; seeing it requires enterprise ownership, billing manager status, organization ownership, or a custom role carrying the 'View Copilot Metrics' permission.[1]
A dashboard that writes a tool's cost as a percentage of payroll defends the tool for as long as the tool looks cheap; the same frame, just as easily, makes a statement about the payroll. What the dashboard does not measure therefore matters: who absorbs the review burden when pull request counts rise, where the cost of defects and rollbacks is booked, and whether saved time turns into pay, into hours, or only into margin. Another explanation is available: the dashboard may be designed as a budgeting tool and may claim no answer to these questions. Even so, something is measurable next quarter — if the same organisation publishes review time and headcount alongside cost per pull request, whose favour the ratio works in becomes visible.[1]