AI-Powered Churn Prediction for the Banking Sector
Challenge
The banking institution had already developed an in-house churn prediction system, but several aspects could be improved to increase operational effectiveness, including:
Reducing false positives. Many customers were flagged as at risk of churning without this actually happening, generating time-consuming interventions for internal resources;
Limited granularity. The system did not distinguish between customers who would fully close their relationship with the bank and those who would simply reduce their use of its services;
Limited interpretability. It was unclear which factors were driving the risk, making it difficult to design retention actions targeted at specific variables;
Short-term forecasts. The limited time horizon did not allow for adequate medium-to-long-term strategies.
The bank therefore needed a more sophisticated AI solution that could not only predict churn with greater accuracy, but also provide interpretive insights to guide concrete action.
Customer retention is a strategic priority for the Swiss banking sector, particularly for institutions offering specialized pension and health-related services.
Our AI Solution
Artificialy developed a churn prediction system based on dual specialized models, each designed to capture a different churn pattern.
The system analyzes historical data from heterogeneous sources, such as digital data (web and mobile banking access), transactional data (volumes, frequency, types of transactions), support interactions (calls, complaints, service requests), and behavioral patterns (credit score, proximity to annual renewal).
This data feeds two complementary predictive models:
Cancellation Churn Model. Calculates the probability that a customer will decide to cancel their subscriptions entirely and stop using banking services. The model provides 3-month forecasts for immediate action on high-risk customers and 6-month forecasts for medium-term preventive strategies.
Transactional Volume Drop Churn Model. Predicts the probability that a customer, without formally closing their account, will drastically reduce their use of services. Specifically, it identifies customers who will carry out at least 90% fewer transactions compared to the same quarter of the previous year. This model captures a phenomenon that is often invisible: customers who gradually drift away without closing the relationship, but who in effect no longer generate value for the bank.
In addition to its predictions, the AI provides an explanation of its choices (explainability), indicating for each individual customer which factors had the greatest influence on the decision, among:
Credit score and recent changes;
Number and type of calls to customer support;
Frequency and type of interactions via digital channels;
Temporal proximity to the annual fee payment;
Anomalous patterns compared to the customer's historical behavior;
And many others
This transparency allows retention teams to design personalized interventions based on the specific risk drivers.
Results
Artificialy's AI solution, tested on a Swiss banking institution specializing in pension services, delivered measurable and significant improvements:
Over 90% reduction in false positives
+14.4% forecast precision
Greater understanding of risk drivers
The solution demonstrates how AI can not only improve predictive accuracy, but transform churn prediction systems from alerting tools into decision intelligence platforms for the proactive management of the customer base.
.png)
