Capture a baseline on day 1 of the validation sprint. With a fast org-wide launch, watch the funnel daily during launch week, then weekly. Report trends rather than single numbers.
Activation funnel
A default-on launch gets the app to everyone quickly. Many of them never start a session. Track each step so you can fix the biggest drop first.
| Step | Definition | Common reason people stall | What helps |
|---|---|---|---|
| Aware | Knows the app exists | Never heard of it, even with a Copilot seat | Launch comms and “try this” prompts in team channels |
| Installed | Downloaded and installed | Unsure it is approved | A clear “approved for use” line in the launch email |
| Opened | Signed in at least once | Sign-in or account confusion | First-hour quickstart in the kickoff |
| First session | Ran one session to completion | Blank prompt box | Starter prompts that run on day one |
| Weekly active | Used the app in four of the last six weeks | No workflow that sticks | Role-based use cases and one automation per team |
Copilot usage reports cover activity across surfaces, but they do not map one-to-one to these steps. For Aware and Installed, use your launch survey and device management inventory. To collect your own usage data, the telemetry key in enterprise managed settings exports OpenTelemetry data to a collector you choose.
The Copilot usage metrics reference includes Copilot app fields under totals_by_copilot_app, such as session_count, prompt_count and used_copilot_app. Use them for Opened, First session and Weekly active.
Parallel work
People get more from the app once they run several sessions at a time. Run work in parallel teaches the habit. Two signals show whether it’s catching on:
| Signal | How to get it |
|---|---|
| App sessions per active user per week | Sum totals_by_copilot_app.session_count per user over seven days in the per-user usage report. A rising median means people start more work in the app. |
| Share of weekly active users who ran two or more sessions in the same hour | Not in the usage reports. Ask in the monthly survey, or count from your own OpenTelemetry export if you collect one. |
Outcome metrics
| Category | Metric | Where to get it |
|---|---|---|
| Launch | Days from kickoff to org-wide enablement (target: 30 or fewer) | Your rollout plan |
| Launch | Share of seat holders who opened the app in the first 14 days after launch | Copilot usage reporting for your organization |
| Launch | Median days from launch to a person’s first session | Copilot usage reporting |
| Adoption | Share of seat holders who are weekly active at day 30, 60 and 90 | Copilot usage reporting |
| Adoption | Share of weekly active users who used the app for something other than code, such as docs, Office files or research | Monthly survey below |
| Adoption | Repositories with custom instructions or skills, and installs of your starter-kit plugin | Code search and your plugin marketplace |
| Adoption | Weekly active users of the app | Copilot usage reporting for your organization |
| Adoption | Repositories with a reviewed .github/github-app.yml |
Code search for the file across your organization |
| Adoption | Active automations per team | Automations list, reviewed in the monthly adoption review |
| Delivery | Median PR cycle time, open to merge | Pull request data from the GitHub API |
| Delivery | Review turnaround time | Pull request review data |
| Quality | Change failure rate or reverts per week | Your incident or deployment tracking |
| Reliability | Crashes and hangs reported per week | Feedback channel and office hours notes |
| Sentiment | “The app saves me time” score, 1 to 5 | Monthly survey below |
Not sure which outcome to lead with? Start from the metric lists the use cases that move each one.
Measure teams, not individuals. Agent activity is a poor proxy for anyone’s performance.
Monthly pulse survey
Keep it under two minutes. Download the survey template.
- In the last month, how often did you use the Copilot app? (Never, monthly, weekly, daily)
- “The app saves me time.” (1 strongly disagree to 5 strongly agree)
- Which workflow saved you the most time? (Free text)
- What stopped you from using it more? (Free text)
- Did the app crash, hang or lose work this month? (Yes or no, with details)
- Besides code, what did you use it for? (Research, docs, Office files, data analysis, automations, triage, presentations, other)
- On a typical day, how many sessions do you have running at the same time? (One, two or three, four or more)
ROI worksheet
The ROI worksheet multiplies hours saved per workflow by the people using it and a loaded hourly cost. Fill in the hours from your survey and time studies, not from vendor claims.
| Input | Example | Source |
|---|---|---|
| Weekly active users | 120 | Usage reporting |
| Hours saved per user per week | 1.5 | Survey median |
| Loaded cost per hour | Your finance number | Finance |
| Seats and other costs per month | Your contract | Procurement |
Show the result as a range. Use the survey’s 25th and 75th percentile for hours saved.
Try it
Change any number. The defaults match the worked example below.
Worked example
These numbers are made up to show the math. Replace every one with your own.
A team has 120 weekly active users. The pulse survey’s 25th percentile is 1 hour saved per week and the 75th is 2 hours. Finance gives a loaded cost of $90 an hour, and the program costs $10,000 a month.
- Hours saved per month: 120 × 1 × 4.33 = about 520 hours, up to 120 × 2 × 4.33 = about 1,039 hours.
- Value per month: 520 × $90 = $46,764, up to 1,039 × $90 = $93,528.
- Net after cost: $36,764 to $83,528 a month.
- Return per dollar: about 4.7x to 9.4x.
Report it as “between 4.7x and 9.4x, based on our own survey”, and say where each input came from. Time saved is not money saved unless the time goes to other work, so pair this with a delivery metric such as PR cycle time.
GitHub’s Copilot impact dashboard also has a potential return on investment section that compares cost and pull request output across adoption cohorts. Use it alongside this worksheet. See Copilot usage metrics.
GitHub’s own adoption and measurement resources
- Rolling out GitHub Copilot at scale: GitHub’s tutorials for planning a rollout, driving adoption and measuring results.
- Driving Copilot adoption in your company: a launch timeline and enablement plan.
- Rolling out the Copilot app to your team: policies, models, MCP servers and a first pilot task.
- Measuring the success of a Copilot trial and measuring the impact of Copilot.
- Viewing Copilot usage and adoption: the usage metrics dashboard, with NDJSON export. Data can lag by up to three days.
- Copilot usage metrics: adoption cohorts, the adoption multiplier and the impact dashboard.
- Measuring the impact of generative AI: a GitHub Well-Architected scenario for building a measurement program.