Customer Churn Prediction

The telecommunications client was losing substantial revenue to customer churn, putting its market position under pressure. It lacked a reliable, automated way to identify at-risk customers before they left.

Templefield Technologies was engaged to build a comprehensive, end-to-end predictive system that identifies customers at risk of churning and surfaces the drivers behind that risk.

• Data ingestion from spreadsheets and SQL databases, with exploratory analysis, cleaning, and validation
• Classification of customers into risk categories using ensembles of decision trees, verified through cross-validation
• Causal analysis linking input data to churn likelihood
• Gradio visualizations displaying churn probabilities and feature importance
• Performance monitoring via MLflow, model serving via FastAPI, and Kubernetes deployment for scalability

The system enabled automated identification of at-risk customers, powering targeted email campaigns that significantly reduced churn and helped the client protect and expand its market share. These prediction-guided campaigns reduced the churn rate by 7%.