A tier-1 bank cut release regression from six weeks to nine days.

The bank's mobile and web platform shipped on a monthly cadence, and every release dragged a six-week regression tail behind it. By the time the suite went green, the next release was already in flight. Quality wasn't the bottleneck anyone admitted to — but it was the one quietly setting the pace of the whole organisation.

The challenge

A regression suite of roughly 3,400 cases had grown by accretion over eight years. Nobody could say with confidence which cases still mapped to real risk and which were duplicates, dead paths, or checks for features that no longer existed. Runs were slow, flaky, and run by hand. A failure meant a half-day of triage to decide whether it was a real defect or just the suite being the suite.

The approach

We embedded a named Velocity Pod — not bodies against a backlog, but a senior QE team that owned the outcome and signed off on it. The work ran in three moves.

• Risk-map the suite. Every case was scored against current product risk. The ones that mapped to nothing were retired; the gaps that mapped to real risk but had no coverage were filled.
• Make runs autonomous. Velocity's agents took over execution on every merge, with self-healing locators that absorbed the UI churn that had been driving most of the flake.
• Move the gate. Regression stopped being an end-of-cycle event and became a continuous signal, with an evidence pack attached to every candidate build.

The result

Nine days, end to end — and most of that is human review, not machine time. The suite is smaller, faster, and every case in it earns its place. The team stopped triaging the test harness and started triaging actual defects, of which they now catch substantially more before release.