The challenge
The client, a fitness training app, already had personalized plans — and ugly numbers: most users generated a plan and stopped opening the app within three days. The team's instinct was that the model was not smart enough and wanted to upgrade the algorithm. We blocked that conclusion first: if you do not know why people quit, a smarter plan just gets ignored more precisely.
How we researched it
UX research ran diary studies and interviews through the first-week experience while the AI side analyzed plan data in parallel. Put together, the truth surfaced: the plans were not bad, but default intensity was idealistic — six training days in week one — and the moment a user missed a single day, the app greeted them with a wall of red overdue sessions. The plan was not the problem. There was no way back after missing a day.
What we did
This is the case where our four practice areas connect. On the AI side we rewrote plan generation with recovery as a first-class requirement: miss a day and the plan shrinks and reschedules itself instead of accumulating debt, and intensity starts conservative, ramping up with completion history. On the UX side we redesigned the return path — after a missed day you see a shortened today-plan and one line that says just pick it up here, not a guilt trip. We validated the whole return path with prototype tests before launch.
Results and the verdict
Day-7 retention and plan completion both rose sharply. The honest verdict: we A/B tested a model-upgrade-only variant and the numbers barely moved. The real unlock was the experience after a missed day. AI decides whether the plan is good; UX decides whether the person comes back — remove either half and this case does not exist.
