Design the evaluation and build the baseline

How will we know if a model is any good? This session answers that before we build anything serious. You'll compare against baselines that need no machine learning, choose a metric that matches the retention team's real decision, split the data honestly, and turn every preparation step into one scikit-learn pipeline that treats new customers exactly like old ones. Then a first logistic regression, a decision about who to call, and an evaluate() function that scores any set of choices. That function is what Earth will use in the labs.