Cohort retention deep-dive for a language learning app

Retention on our language learning app looks fine in aggregate but we suspect it is hiding two very different stories: exam-preppers who churn the day after their test, and hobbyists who stick around for years. Before we redesign onboarding around the wrong persona, we want someone to actually segment the base and prove or disprove this. You will get read access to our Postgres warehouse (event data for ~800k users over 3 years) and a list of the product questions we are wrestling with. Expected output: cohort curves cut by acquisition channel, stated goal, and first-week behavior; a survival analysis of the top segments; and a working session where you walk our product team through what the data supports and what it does not. SQL and Python notebooks stay with us. We budget this at 10-15 hours a week for about three weeks. Our head of product is your main contact and responds fast. Curiosity about the why behind the numbers matters more here than fancy modeling.