Predicting Customer Attrition: Churn Prevention for a Leading P&C Insurance Company .
Overview.
A leading property and casualty (P&C) insurance company aimed to predict customer attrition, both mid-term cancellations and end-term non-renewals, to increase its persistency ratio. Coforge implemented advanced machine learning techniques, including feature engineering, survival analysis, and ensemble learning models, to accurately identify high-risk customers and improve retention rates.
The client faced significant challenges in predicting customer attrition:
Customer Attrition Prediction: Needed to predict both mid-term cancellations and end-term non-renewals.
Increasing Persistency Ratio: Aimed to increase the persistency ratio by identifying and retaining high-risk customers.
Solution.
Coforge implemented a comprehensive churn prevention solution:
Feature Engineering: Combined variables from socio-demographic data, customer-company interactions, product and servicing details, claims, and competing product offerings.
Feature Selection: Used the Boruta algorithm for effective feature selection.
Survival Analysis: Applied survival analysis techniques to address attrition probability and timing.
Separate Modeling: Modeled cancellations and non-renewals separately for more accurate predictions.
Ensemble Learning Model: Developed an ensemble learning model using various techniques, including Random Forest, AdaBoost, CatBoost, XGBoost, and LightGBM.
Key Highlights:
Coforge's solution delivered significant value to the client's operations:
Early Flagging: Identified high-probability and high-value customers early.
High Model Accuracy: Achieved a high model accuracy of approximately 92%.
The Impact.
Customer Identification Early flagging of high-risk customers
