Salmon is a fast-growing mobile bank. As the product accelerated, our regression load grew faster than the team could test — and release fear set in. We tried to fix it by rebuilding the QA process. We broke it twice before anything held.
This is the honest version, not a polished case study. Both rebuilds looked reasonable on paper: we hired, we changed how the team worked, we reshaped the workflow. Both times the classic approach collapsed under a product that changes faster than tests can keep up. We'll show exactly where it broke, what it cost us, and what we missed each time.
The third attempt started working. We have E2E infrastructure on mobile, devs automating against it. We don't chase a full pyramid or pure E2E — we split each story into layers, from cheap snapshots to long E2E flows. We test stories end to end, mobile through backend, not each side alone, learning not to duplicate checks. It's not perfect, but for the first time in four years we see how to scale this.