Customer Churn Prediction Platform
Impact$2M+ annual savings
Accuracy85%
Retention Lift30%
ROI$2M+
Overview
Developed a churn prediction platform for a telecom provider to identify customers at risk of leaving. The system provides churn scores, key drivers, and recommended retention actions for each customer.
The Challenge
The telecom was losing thousands of customers monthly with no way to predict or prevent churn until cancellation requests came in.
The Solution
Built gradient boosting models trained on usage patterns, billing history, and support interactions. Created a real-time scoring pipeline that updates daily and triggers retention workflows.
Key Results
85% accuracy in predicting 30-day churn
30% improvement in retention campaign effectiveness
$2M+ annual savings in reduced churn
Tech Stack
PythonLightGBMApache SparkAirflowRedshiftLookerAWS Lambda