Customer Churn Prediction and Retention ROI Dashboard Using Explainable ML

Churn TIMING, not only churn risk: a churn classifier says who is likely to leave but not when, and a retention budget needs both. The study treats each customer's tenure as a survival time with churn as the event - Kaplan-Meier curves by contract and payment method with log-rank tests, a Cox proportional-hazards model with the Schoenfeld check, a Cox model stratified by contract, a random survival forest and a gradient-boosted Cox model, all 5-fold cross-fitted and compared by concordance (Harrell, Uno), time-dependent AUC and the integrated Brier score. Each customer's survival curve gives an expected churn month and a 36-month customer lifetime value (CLV); four retention offers with stated, ASSUMED effects are valued and a CLV-weighted budget optimiser (multiple-choice knapsack) is compared with 'top-risk first' and random targeting. A small local LLM drafts retention e-mails per segment behind a rubric gate with a template fallback. The IBM Telco Customer Churn data is a FICTIONAL sample company created by IBM: the results demonstrate the method and are not evidence about real customers.

Category: Business analytics

Master of Business Administration Project MBA04 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.