AI Impact

Set it up

1. Open Developers › Data sources › GitHub (admin or owner).
2. Choose Connect GitHub, top right. GitHub asks you to pick the organization and the repositories to install the App on, and to authorize it. You must be an admin of that GitHub organization. If you are not, GitHub sends an install request to an organization owner; connect again after they approve it.
3. Back in Token Meter the first sync is queued. The connection card shows the account, status, last sync and how many repositories are included.
4. Under Repositories, include or exclude each repository. Only included repositories are read. Forks and archived repositories start excluded.
5. Under Git authors, check that each GitHub login is matched to a workspace member (see below).

To add or remove repositories later, use Reconnect or change repositories. Sync now queues a sync straight away. Disconnect stops syncing and removes the stored credential.

What a sync reads

• Syncs run about once an hour. The first sync of a repository reads the last 30 days, and the past year is then filled in behind it over the following hours; later syncs re-read the last 3 days, because pull requests merge late and history can be rewritten.
• Commits on each repository's default branch (merge commits are skipped), with their date, author, line counts and AI markers.
• Pull requests in any state (merged, open, closed), with dates, size, commit count, reviews by people other than the author, review threads and revert links.
• Jira-style issue keys (such as PAY-123) found in pull request titles, branch names and commit messages, so tickets can be linked to pull requests.
• 14 days after a pull request merges, line survival measured with GitHub's blame (no repository clone).

Matching authors to people

At each sync, a GitHub login is matched to a workspace member whose email appears on that login's commits. When someone commits with a personal or noreply address, map the login by hand under Git authors: enter the member's email and choose Map. A manual mapping is never overwritten by a sync. Bot and agent accounts (for example copilot-swe-agent[bot]) are not people and cannot be mapped.

Work by an unmatched author shows as Unassigned on the team charts.

Set it up

1. Each Jira project is mapped to a Token Meter team, so create your teams first (Settings › Team › Teams). The Jira page lists every existing team with its slug.
2. In Jira, find the story points field id: Settings › Issues › Custom fields › Story points; the id (such as customfield_10016) is in the field's URL.
3. Create an API token for a dedicated read-only integration account that can browse projects and issues in the projects you map.
4. Open Developers › Data sources › Jira (admin or owner) and fill in the form below.
5. Choose Save Jira connection, then Sync now. The page shows the number of imported events, the last sync and any error.

Later changes (a new token, mappings, story points field) go under Rotate credentials or change project mappings. Disconnect stops syncing; imported issues are kept. To cut off access completely, also revoke the token in your Atlassian account.

What a sync reads

• Syncs run about once an hour. Each sync reads issues in the mapped projects changed since the last sync (with a 2-day overlap); the first sync reads the lookback window (365 days by default), and so does the next sync after the lookback is changed.
• Per issue: key, project, issue type, parent and epic, status category, priority, story points, and created, updated and resolved dates.
• Issue titles, descriptions and comments are not requested or stored. No assignee or reporter is read.
• The story points field must exist and be a number field. If it is not, the sync fails and the error lists likely candidate field ids.

Done without a resolution

Many team-managed Jira projects move issues to Done without ever setting a resolution. AI Impact treats an issue as resolved at its resolution date or, when there is none, at the time it entered a Done-category status. An issue outside the Done category with no resolution date is unresolved.

Repositories

• By repository: merged pull requests, AI-authored pull requests, AI spend (AIScale) and spend per merged or per AI-authored pull request. Select a repository for lines merged, spend with no committed code, estimated spend across all tools, tools and models, and review coverage.
• Quality compares AI-authored and human pull requests: reviewed before merge, reviews per pull request, changes requested, commits after first review, rework rate, 14-day churn, time to first review and reverted. Review data exists only for pull requests synced since review collection started. Recent pull requests lists pull requests by team, not by author; the number opens the pull request's detail page.
• The pull request detail page also shows Review, Did the code stay? (line survival, and whether it was reverted) and Linked Jira tickets.

Delivery

• Cost per PR: cost per AI-authored pull request by team, weekly or monthly, with a trend verdict once a team has 8 weeks of history (first four weeks against the last four). Without GitHub, the tab uses the pull requests, commits and lines the AI tools themselves report. It also shows spend over time, AI cost by Jira epic or project, story points against spend, Before vs after AI and Spend by model.
• Jira: By project (resolved issues, story points, bugs created, Est. AI spend), Bugs (created, fixed, open now, median lead time from created to resolved), By epic and AI cost per ticket.
• Recall: estimated context tokens avoided by Agent Recall; see Router.

Team

Org adoption rate, daily active AI users, power users (AI use on at least 75% of workdays), average cost per active user and usage by role. Admins set a target per team with Edit targets: the share of the team's roster that should be active, over a window of 7 to 365 days. A team below its target raises a warning alert, checked daily, which clears when the team is back on target.