Predicting B2B Customer Churn and Measuring the Impact of Machine Learning-Based Retention Strategies

Victória Emanuela Alves Oliveira, Amanda Cristina da Costa Guimarães, Arthur Rodrigues Soares de Quadros, Reynold Navarro Mazo, Rickson Livio de Souza Gaspar, Alessandro Garcia Vieira, Wladmir Cardoso Brandão

Abstract: Acquiring new customers often costs five times more than retaining existing ones. Customer churn significantly threatens B2B companies, causing revenue loss and reduced market share. Analyzing historical customer data, including frequency on product usage, allow us to predict churn and implement timely retention strategies to prevent this loss. We propose using Support Vector Machines (SVMs) to predict at-risk customers while retraining it, if necessary. By monitoring its recall over 15-day periods, we retrain the model if its recall on new data falls below 60%. Our research focuses on feature selection to identify key churn factors. Our experiments show that when constantly retraining the model, we avoid accuracy loss by updating the customer’s context, providing valuable insights on how to reduce churn rates and increase revenue.