Churn Modelling

The goal of a churn modelling project is to identify which customers are likely to leave before they do. With that information, a business can decide whether to intervene, who to target, and what action to take. Without a model, retention efforts are based on guesswork.

The project typically starts with collecting and preparing customer data: usage history, transaction records, contract information, and complaints. A model is then trained to predict the probability that each customer will leave within a given period. This requires defining what "churn" means for the specific business, which is not always straightforward. The final model is evaluated on a held-out dataset and calibrated to the business's needs.

Common approaches include logistic regression, gradient boosting, and survival analysis. Python and R are the main languages used. Tools like scikit-learn, XGBoost, and pandas are standard. The choice of model depends on data availability and how the output will be used. Predicted probabilities matter as much as model accuracy, since they feed directly into decision rules.

Churn modelling projects appear in telecoms, financial services, and insurance in the Netherlands, as well as in subscription software companies. The project connects to Data Scientist, Data Analyst, and Risk Manager roles. In financial services, churn modelling overlaps with customer lifetime value estimation and credit risk work.