SmartChurn: Retention AI

Business Problem

A telecom subscription service was facing increasing churn but lacked visibility into who was leaving and why.

Solution Approach

Engineered a robust pipeline using aggregated usage logs and support ticket sentiment. Trained an XGBoost classifier with SMOTE for class imbalance, achieving 0.89 F1-score.

Business Impact

Identified $2.3M in at-risk revenue annually; empowered retention teams to target interventions with 3x higher precision.

Methodology Note

"Risk calculation based on a high-value segment ARR of $10M with a historical 23% churn rate. The model targets the top decile of risk with 90% recall, identifying $2.3M in preventable revenue loss annually."

Key Features

• • Interactive Streamlit dashboard
• • Feature importance and SHAP analysis
• • Batch and real-time predictions
• • Model comparison and ROC-AUC metrics