AI impact
Compare pull requests and deployments made with AI coding tools with the rest of your work, from merge share to gate pass rates and deployment-based DORA.
Availability
The AI impact page (Insights > AI impact) puts AI-assisted work next to the rest of your work: how much of what you merge is AI-assisted, whether it passes your gates as often, how fast it moves, and whether it fails in production more often. Every number is for the whole organization or one repository. There are no per-person numbers.
Choose the period (7, 30 or 90 days) and a repository at the top of the page. Both are kept in the page address, so you can share a link to the same view.
Three groups, never two
Prodgator classifies each pull request and each finished deployment:
- AI-assisted: at least one signal matched, such as a commit trailer naming an AI tool, an AI agent account, or one of your AI labels.
- Other: classified, and no signal matched.
- Not classified: Prodgator has no classification for it, for example a pull request merged before your plan included AI impact. These are counted on their own and never added to "Other".
Pull requests opened by bots that are not AI agents (Dependabot, Renovate and similar apps) are left out of both groups.
What the page shows
| Section | What it shows | Where it comes from |
|---|---|---|
| AI-assisted share | AI-assisted merged pull requests as a share of all classified merged pull requests, with the number not classified | Classified pull requests, by the day they merged |
| Pull request gate first pass | How often a pull request passed its gate on the first evaluation that reached a result | Pull request gate evaluations |
| Change failure rate | Failed or rolled back deployments as a share of finished deployments | Deployments to protected environments |
| Merged pull requests per week | AI-assisted and other merged pull requests per week (weeks start on Monday, UTC) | Classified pull requests |
| Lead time | Median time from a pull request opening to the deployment that shipped it, successful deployments only | Deployments to protected environments and the pull requests they shipped |
| Cycle time | Median and 75th percentile time from ready for review to merge | Pull request events (GitHub, GitLab and Azure Repos) |
| Gates, findings and CI | Gate first-pass rates, overrides and bypasses per 100 gated changes, deployments per day, rollback rate, time to restore, review time and coverage, open findings on merged heads, CI failure rate and top causes | Gate evaluations, deployments, pull request events, security scans and pipeline runs |
| How we classify | AI-assisted merged pull requests by signal and by tool | Classified pull requests |
Every chart has Show as table for the same numbers.
Deployment numbers and protected environments
The DORA numbers on this page are deployment-based: they count deployments to protected environments only, deployment environments that a gate covers and does not exclude. An environment's name does not matter. Past days are counted by your gates as they are today. This is not the same as the run-based DORA metrics on the Analytics page, which are counted from pipeline runs.
Until a gate protects a deployment environment, every deployment number is empty and the page links to Gates. A deployment is AI-assisted when it ships an AI-assisted pull request or a commit with an AI trailer.
Findings
Findings are the open security findings in the scans of each pull request's merged head commit: what the gate saw at merge time. They are not only the findings the pull request introduced. Critical and high counts the critical and high severity ones.
Small groups
A median needs at least 5 samples in its group. With fewer, the page shows Not enough data instead of a number: a median of two pull requests is noise, and a small group in one repository can point at one person. Rates are shown with any number of samples.
When numbers appear
A daily rollup at 02:15 UTC counts each day on the following day, so today is not included. Before the first rollup covers the period, the page says when the numbers will appear.
- Cycle time and review times exist for the last 35 days, the time Prodgator keeps pull request events.
- Pull request overrides and bypasses exist for the last 30 days.
- Bitbucket Cloud pull requests count toward the share, gates and findings but have no cycle or review times.
When nothing is classified
If no merged pull request or finished deployment in the period was classified, the page shows an empty state instead of numbers. Check that:
- your plan includes AI impact (Business or Enterprise),
- your repositories are connected and pull requests have merged since then,
- the settings in Admin Console > AI > AI-assisted changes match how your team marks AI work, for example the labels you use. See AI-assisted changes.
Act on it
To ask more of AI-assisted changes, such as an extra approval or a label that says a person checked the change, add the AI-assisted changes rule to a pull request or release policy.
Ask Prodgator
When you ask from this page, the assistant receives the numbers on screen (counts and rates only, no repository or tool names). It can also read the same view for any period or repository with its AI impact tool, so you can ask questions such as "Do AI-assisted pull requests fail CI more often than the rest this quarter?".