From no QA function to a weekly release
Two years of shipping with no dedicated testing had started to show. We stood up an embedded team without making delivery slower on the way.
- Sector
- Team
- Duration
- Service
The situation
Eighteen engineers, no testers, and a product that had outgrown the founders' ability to check it by hand before a release. Nothing was on fire — things were quietly getting worse, which is the harder case to make internally.
The instinct was to buy a testing team — six people, arriving at once. At that size and that pace, that usually produces a bottleneck rather than a safety net: the testers become a stage everything queues behind, and delivery slows exactly when the company can least afford it.
What we did
Started with two people, not six
A quality engineer and a lead, embedded in the existing teams rather than sitting beside them. The brief was to make the engineers better at testing before adding anyone whose job was to test on their behalf.
Automated the three journeys that carried the revenue
Listing, checkout, and payout. Everything else stayed manual for the first month on purpose — a broad, shallow suite built at that speed becomes the flaky one we get called in to delete somewhere else.
Put the release on a schedule the team chose
Weekly, on a day they picked, with a rollback rehearsed before the first one went out. A cadence people choose is a cadence they keep; one handed to them survives until the first awkward week.
Grew the team to fit delivery, not a headcount plan
Six months in it is four people, two of them the company's own hires, whom we interviewed and onboarded. The engagement is designed to shrink as the internal function grows.
The outcome
The first fully automated release went out in week eleven, and the weekly cadence has held since. Merge to production is a day, down from six.
Defects reaching production are down 64% against the two quarters before we started. The team's own measure is that nobody checks the marketplace by hand on a Friday afternoon any more.
What we'd do differently
We should have refused the first scope faster. The original ask was a six-person managed team from month one, and we took two months to talk them out of it instead of two weeks. Every week that conversation ran was a week the two people they actually needed were not there.
- 11 weeks
- To the first fully automated release
- 1 day
- Merge to production
- from 6 days
- −64%
- Defects reaching production