Case Study FinTech Predictive Churn & Customer Lifetime Value ML Model (UK)

Project Overview

Gradient-boosted machine learning pipeline trained on 40M+ banking transactions to identify early customer churn signals 90 days before account abandonment.

Client & Location

Sterling Digital Bank Ltd
Location: London, UK
Industry: Neo-Banking & FinTech

Challenges

• Account attrition eroding deposit base with no early warning indicators.
• Massive tabular transactional dataset with significant class imbalance.
• Strict European banking regulatory compliance prohibiting black-box unexplainable AI models.

Solution

Engineered an explainable XGBoost model pipeline utilizing SHAP (SHapley Additive exPlanations) values to output human-readable churn drivers for relationship banking officers.

Technologies Used

• Python 3.11, XGBoost & LightGBM
• SHAP Explainability Framework
• Apache Spark on AWS EMR
• MLflow Model Registry & Feast Feature Store

Results & Business Impact

• Predicted customer churn with an AUC-ROC score of 0.941.
• Saved £4.2M in annual customer deposits via automated proactive retention triggers.
• Full regulatory audit approval achieved with complete model explainability documentation.