Governing_Predictive_Customer_Targeting_Bank_Telemarketing

This book chapter is a decision-oriented re-evaluation of predictive models (CART, CHAID, C5.0, and logistic regression) for targeting customers in bank telemarketing campaigns for term deposits. Using the UCI Bank Marketing dataset, it shows that comparing models only on default classification thresholds can be misleading. When models are compared at the same contact-list size (matching campaign capacity), their ability to capture subscribers becomes much more similar. The chapter also demonstrates that performance changes significantly under chronological (time-ordered) validation because subscription rates and economic conditions drift over time. It emphasizes practical governance: define campaign capacity first, evaluate models at matched operating points, validate temporally, and keep implementation uncertainty visible. The work supports more accountable and responsible use of predictive targeting systems in banking.