Product Manager
Two years ago I shipped a new onboarding flow for our mobile app. I had user interviews, prototypes, the whole thing. We rolled it out to a hundred percent of new users on day one. Day-one activation dropped fifteen percent in the first week. The flow was well-designed in isolation, but it assumed users knew what we did, which our old flow had been quietly teaching. I should have shipped behind a feature flag and ramped to ten percent first. We rolled back and rebuilt the introduction sequence over the next three weeks. Activation came back, but the original launch cost us roughly twelve hundred users we never recovered. What I do differently now is treat any change to onboarding as an experiment by default, never a launch. Even when I think the new version is obviously better, especially when I think that.