In highly competitive consumer credit markets, understanding customer churn is critical. This project focused on predicting early loan repayment by integrating multi source behavioral data into a single, explainable machine learning model to support proactive retention strategies.

A leading Italian financial company specializing in consumer credit, operating across multiple digital and offline customer touchpoints.

The client needed to better understand churn likelihood in a complex environment characterized by fragmented data sources and multichannel interactions.

Key challenges included limited visibility into early digital signals preceding loan closure, lack of integration between online and offline touchpoints, and the need to process large volumes of raw data in a scalable and structured way.

Custom machine learning models developed within a cloud-based data infrastructure, combining advanced analytics, feature engineering, and explainable AI techniques.

The initiative involved the design and development of a custom anti-churn predictive model.

Data from multiple sources, including digital behaviors, offline interactions, marketing data, and sociodemographic information, were collected and prepared. Extensive exploratory analyses were performed, followed by the creation of over 1,800 predictive features.

Multiple machine learning algorithms were tested and benchmarked to identify the most effective approach, supported by a scalable cloud environment for analysis and model development.

  • Improved churn prediction accuracy through integrated, multi-source data
  • Clearer understanding of churn drivers thanks to explainable model outcomes
  • Churn scores shared with data activation systems
  • Enablement of targeted retention actions, such as personalized campaigns
  • Increased data quality, awareness, and governance maturity