You are building a supervised learning model for a consumer genealogy platform to predict whether a user will start a paid subscription after engaging with records, family trees, and discovery features. The marketing and product teams want a user-level propensity score they can use to prioritize outreach and personalize the experience.
How would you approach building this model to predict user behavior, from defining the target and features through model selection, training, and evaluation? What tradeoffs would you consider before putting it into production?