Deposit Churn Prediction: Predictive analytics to retain depositors and protect banking revenue
Project Overview
This project focused on building a predictive analytics solution to identify bank customers at risk of reducing or closing their deposit accounts. By analysing transactional behaviour, account activity, and customer attributes, the model enables banks to detect early churn signals and engage customers proactively. The solution helps retention and marketing teams prioritise outreach, protect deposit balances, and strengthen long-term customer relationships.
Problem Statement
Banks traditionally relied on reactive and manual methods to identify deposit churn, often discovering disengagement only after balances had already declined. The absence of predictive insights limited the ability to intervene early, resulting in preventable revenue loss and reduced customer lifetime value. Retention efforts were not prioritised effectively, leading to inefficient engagement and missed opportunities with high-value customers.
Key Findings
- Early Behavioural Churn Signals: Declining account balances, reduced transaction frequency, and extended inactivity beyond 90 days consistently emerged as strong early indicators of deposit attrition.
- Hidden Risk Among High-Value Customers: Premium depositors often showed subtle declines in engagement well before reducing or closing accounts, making proactive monitoring essential for protecting high-value relationships.
- Effectiveness of Personalised Retention: Targeted, data-driven retention offers significantly outperformed broad engagement campaigns, delivering higher response rates and improved long-term customer retention.
Implemented Solution
A machine learning-driven churn prediction framework was developed to enable early detection and proactive retention:
-
Churn Prediction Pipeline:
Built a robust prediction model to calculate churn probability for deposit customers, using historical transaction data, behavioural signals, and customer profiles, to enable early and reliable churn identification.
-
Advanced Feature Engineering:
Engineered features capturing balance trends, transaction patterns, inactivity periods, and customer segments to improve predictive accuracy and insight depth, while enhancing model interpretability for business users.
-
Real-Time Monitoring Dashboards:
Integrated automated churn scoring into interactive dashboards, providing teams with real-time visibility into at-risk customers and churn trends, supporting faster decisions and proactive intervention.
-
Retention Enablement:
Delivered actionable insights to marketing and relationship teams, enabling targeted retention campaigns focused on customers with the highest churn risk and value, maximising impact while optimising retention effort.
Results
The deposit churn prediction model achieved 89% accuracy, giving banks a reliable early-warning system for customer disengagement. Proactive engagement strategies driven by predictive insights reduced deposit attrition by 18%, while improving customer lifetime value through earlier intervention. Automation of churn detection also enhanced operational efficiency, allowing teams to shift from reactive responses to structured, data-driven retention management.