Measure Value

Are people meaningfully adopting AI?

Licence count is not adoption. Follow each stage from access to measured value, find where people drop out, and import usage exports from GitHub Copilot, Microsoft 365 Copilot or your AI gateway.

Start from:

Import a usage export

A CSV with a user column and a date column: a per-user activity export (one row per user per active day), a GitHub Copilot seat list (assignee_login, last_activity_at) or a Microsoft 365 Copilot usage report. The file is read in your browser and never uploaded. Download a sample file.

Dates like 05/10/2026 are

Adoption funnel

Enter what you have; leave a stage empty if you cannot measure it yet. Mark later stages from evidence, not estimates.

Eligible usersPeople whose work AI could help: licensed, invited or in scopeAI enabledHave access: a seat, an account, the tool installedActivatedUsed it at least onceWeekly activeUse it in a typical weekRepeat usersActive in several different weeks, not just onceWorkflow integratedUse it inside a defined workflow or process, not ad hocMeasured valueA measured outcome: time saved, quality or revenue, with a baseline

Import a usage export or enter your counts to see where people drop out between getting access and getting measurable value. Or start from an example above.

Assumptions & sources

Stages: each stage should be a subset of the one before, counted over the same period. Percentages are of eligible users and of the previous measured stage. The biggest drop-off is the step that loses the most people; it is counted in people rather than percentages because later stages are small. There is no composite adoption score: each stage is shown as measured.

Usage files: weekly figures use Monday-to-Sunday weeks (UTC), and part-weeks at either end of the file are left out of the average so they do not drag it down. For a seat list or last-activity report, the tool reports how many people were active in the last 7, 28 and 90 days before the newest date in the file, and counts listed people with no activity date as never used.

Dates: numeric dates are read day first or month first by checking the file (a first part over 12 means day first). When every date fits both, the setting above decides.

Activity shows use, not value. Pair it with workflow and outcome measures, such as the engineering productivity comparison and the developer experience survey, before drawing conclusions.

Questions

How do I measure AI adoption in my organisation?

Track people through stages rather than counting licences: eligible, given access, tried it once, use it weekly, use it across several weeks, use it inside a defined workflow, and getting a measured outcome. The biggest drop between two stages is usually the best place to act.

How do I measure GitHub Copilot adoption?

Export the Copilot seat list (it includes each person's last activity date) or a per-user activity export, and import it here. The tool reports how many people were active in the last 7, 28 and 90 days, how many seats were never used, and fills the adoption funnel for you.

What is a good weekly active rate for AI tools?

There is no universal benchmark, and it depends on how many roles the tool really suits. Track the trend after rollout and the drop-off between stages, and compare it with the adoption your business case assumed.

Is my usage file uploaded?

No. The file is read in your browser and never sent to Learn AI Engineering. Only the funnel counts are kept in the page address so you can share the result.