Frame a Home-Page Recommendation System End to End

You are the ML lead for a subscription streaming platform with 40 million subscribers. Your team owns the home-page recommendation system. Currently, the page shows a mix of editorial rows ("Popular this week," "New releases") and a lightly personalized row based on a 3-year-old collaborative filtering model.

The VP of Product gives you a loosely defined mandate: "Make the home page better." Walk through five framing decisions in sequence. Do not jump ahead — the answer to each decision should constrain the next.

1. What business objective should this system serve?
2. Is ML the right tool? Name a non-ML baseline you would still compare against.
3. Complete the framing stack: product outcome, model goal, decision policy.
4. What proxy label would you choose, and what is its biggest downside?
5. What would you monitor in production — one business KPI, one model metric, one guardrail?

Follow-up ladder

1. Rung 1: You discover the editorial rows outperform the personalized row on the primary KPI. What does this tell you about the current ML model, and how does it change your plan?
2. Rung 2: Your model goes live. After 4 weeks, watch hours per subscriber increase 6%, but 30-day churn also increases 2%. How do you interpret this, and what do you do?
3. Rung 3: The recommendation system starts serving a small but growing share of traffic on a new mobile app that has no historical viewing data for its users. How does this change the model goal and the decision policy?