Question
Design a churn-prediction pipeline for a B2C subscription product with ~20M active subscribers. The model scores every subscriber daily for likelihood of cancelling within the next 30 days, and the scores feed retention campaigns (discounts, outreach). Features come from billing events, product usage logs, and support tickets. The business wants the model retrained regularly without leakage, scores delivered to the marketing system each morning, and an explanation of why a given user is flagged (for the retention team). This is a batch scoring problem — no sub-second serving requirement.
Clarify scale and constraints first. Propose a clean component breakdown, then go deep on the hard parts — data model, bottlenecks, consistency, failure modes — and name the trade-offs you are making.