Predictive Test Selection

Predictive Test Selection runs only the tests relevant to a change and skips the rest — recording why every test was selected or skipped.

Try Predictive Test Selection Free Request a Demo

Feature overview

Predictive Test Selection

Predictive Test Selection uses a machine-learning model that learns from your organization's own build history to run only the tests relevant to a given change. During each build, the Develocity agent fingerprints what the change touched, scores every candidate test against the model, runs the selected subset of tests, and records each test's outcome and why each test was selected or skipped in its Build Scan. Because test execution time scales with the size of the change rather than the size of the overall test suite, developer feedback remains fast — even as AI multiplies the tests your pipelines have to run.

Works great with

Run only the tests needed to verify a change

On each build the Develocity agent fingerprints inputs, identifying what changed, and scores tests against a prediction model to determine which need to be run to verify the change.
The fingerprint reuses the task-input snapshot Build Cache already computes, so it catches toolchain bumps, generated files, and environment shifts a tool like Git-diff never sees.
The model actively learns from your Build Scan history so predictions adapt as your projects evolve.

Predictive Test Selection view for a single test task — mean duration 43 min 7 sec (41% faster with test selection enabled), a duration trend with mean, median, and percentile bands, and recent builds listing not-selected test classes and serial test time saved per build.

Tune the coverage-speed trade-off

Recently failed or flaky tests always run, regardless of selection score; the model errs toward selection when uncertain.
The Conservative, Standard, and Fast profiles set how aggressively the model selects tests, so the platform team owns the trade-off between time saved and test coverage.
See the savings and the catch-rate before you skip anything — the Simulator replays the model against your Build Scan history.

Build Scan tests table showing one test class kept and run with flaky and passed methods while three sibling classes are marked NOT SELECTED by Predictive Test Selection.

Observability

Every select-or-skip decision, visible per build

Every test is reflected in the Build Scan with the reason why it was selected or skipped.
Trace any one test across historical builds to see how often the model skipped it and whether it passed the times it ran — no guesswork about its behavior.
Each build records how much test time was saved, so you can measure the model's cumulative impact on test wall-clock time.

A single skipped test class in a Build Scan, showing its not-selected reason and estimated time saved.

Analytics

Track realized savings and adoption across every project

Review realized savings by build tool, project, and CI versus local.
Prioritize roll-out for projects where the test phase dominates build time, using Develocity Analytics to find them.
Query your analytics data in plain language through an AI assistant connected via Develocity's MCP Servers.

Develocity Analytics Predictive Test Selection dashboard showing total projects, total not-selected test classes, total serial savings and wall-clock savings, with per-build-tool breakdowns.

The model comes to your data, your data stays put

The model is bundled as part of Develocity and every prediction it makes stays local.
Build data never leaves your Develocity instance, so Predictive Test Selection remains compliant with the data-residency and isolation rules that govern your organization.
Predictive Test Selection does not access external systems like source-control to make predictions; the only input is the change fingerprint and list of candidate tests computed inside your build.