The Problem

Telecom businesses lose millions annually to churn without knowing which customers are at risk or why — until it's too late. Batch reports and spreadsheets don't give retention teams the real-time, actionable intelligence they need.

What I Built

• Trained an XGBoost churn classifier achieving ROC-AUC 0.84 (5-fold CV: 0.84 ± 0.01) with SHAP feature explanations — surfacing that month-to-month customers churn at 42.7% vs. 2.8% on two-year contracts, and first-year customers churn at 47.7% vs. 9.5% after year four
• Quantified $1.67M annual revenue at risk across the churned customer base, enabling data-driven prioritization of retention campaigns
• Built a five-page interactive React dashboard — overview KPIs, cohort retention curves, segment analysis by contract/service/payment method, customer-level risk table, and model insights with SHAP waterfall charts

Impact

Converts a black-box churn model into an actionable retention intelligence platform — business teams can slice, filter, and act on customer risk in real time without touching a data warehouse.