AI Adoption
AI Adoption measures how much the team actually uses artificial intelligence tools when programming, and what effect this has on the work performed. Instead of relying on declarations or on the mere fact of holding licenses, Q247 provides measurable data:
- how many Calories and Increments (Q247's own units for measuring development effort) were produced with the help of AI, compared to work without it,
- how many people actually use AI, not just have access to it,
- how much it costs, broken down by projects and teams,
- whether changes made with the help of AI stay in the code as well as other changes.
This lets management and team leads assess the return on investment in AI tools based on actual usage, not declarations.
Where it appears in the application
The AI Adoption panel appears on four views, with different scope and breakdown:
| View | Data scope | List breakdown when expanded |
|---|---|---|
| Overview | the whole organization | projects |
| Teams | the whole organization | teams |
| Team | a single team | participants of that team |
| Project | a single project | participants of that project |
The panel's visibility is controlled by a separate toggle in Configuration ("Show AI adoption").
Indicators
- % adoption:
AI calories ÷ (AI calories + non-AI calories)in the selected period, that is what share of programming effort was produced with the help of AI. - AI cal: the sum of calories generated with the help of AI.
- Tokens: the sum of input tokens, output tokens, and those that write a new fragment to the AI tool's cache, while tokens only read from that cache are deliberately excluded. There are usually very many of them, they are cheap, and including them would artificially inflate the result.
- AI spend: the estimated cost of AI usage in this period, in USD.
- Cal/$: how many AI calories were generated per dollar spent:
AI calories ÷ AI spend. - R30 retention (AI / non-AI): what percentage of the lines of code added by AI commits (separately: by the remaining commits) survived 30 days without being overwritten or deleted, a weighted average by the number of added lines. It shows whether changes made with the help of AI stay in the code as well as other changes. The full description of the retention metric is on the Retention page.
- Calories per commit (AI / non-AI): the average number of calories per commit, separately for AI and the rest.
- Active AI users / total: how many committers had at least one AI commit in this period, against all committers.
When broken down by project, tokens and cost are allocated in proportion to how many AI calories a given user generated in a given project on a given day. Example: if a person generated 60% of their AI calories in Project A and 40% in Project B on a given day, their tokens and cost for that day are split in the same proportion, 60/40. This way, the totals for all projects add up exactly to the organization's daily usage, nothing is lost or duplicated. When broken down by teams, it is calculated differently: each member's full usage goes to their team, without splitting between teams.
Data sources
AI Adoption combines two independent sources.
Classification of commit content. Q247 analyzes the content of every commit looking for trailers indicating AI involvement, e.g. "Co-Authored-By: Claude". This works on data Q247 already collects. It requires no additional integration. This is the basis for calculating: % adoption, AI cal, R30 retention (AI/non-AI), and Calories per commit. By default, patterns are recognized for popular tools, e.g. Claude Code, GitHub Copilot, Cursor, Aider, Devin, or OpenAI Codex. The list is configurable, so these are examples, not a closed set.
OpenTelemetry telemetry. Only Tokens, AI spend, and Cal/$ come from actual AI tool telemetry, sent to Q247 by the Enterprise Plugin installed in the organization's infrastructure. Q247's standard for collecting this data is OpenTelemetry: the mechanism is designed for any AI tool emitting data in this format, and adding another provider requires only a separate module that reads its data, with no changes to the core of the integration. Currently, the only implemented and supported source is Claude Code. This source requires prior configuration, described in AI tool logs. Without it, the remaining indicators (based on commit classification) still work, but these three are not available.
Both sources are combined at the user level, which requires mapping the account used by Claude Code (usually the email configured locally on the developer's computer) to that person's account in Q247. These addresses are not always the same. Account mapping is described on the Users And Teams page, and the configuration of the telemetry itself is in Configuration.