Previsão e interpretação de churn: integrando análise causal e aprendizado de máquina para estratégias de retenção efetivas

Abstract

Globalization and the widespread use of the internet have transformed the relationship between consumers and companies, establishing a direct and active interaction between them. In this scenario, deeply understanding the customer lifecycle is vital to maintaining the operational and financial stability of organizations, with a keen focus on factors that promote customer satisfaction and loyalty. Faced with the issue of churn – which reflects customer loss – various industries face challenges that directly impact their profitability and sustainability. Thus, this research aims to develop a tool that enhances the predictive modeling of churn, enriching it with causal analysis to not only predict more accurately but also offer clear interpretations of the reasons for customer loss. Using the IBM Telco Customer churn dataset, version 11.1.3, as empirical support, the study seeks to identify influential variables in churn and evaluate effective retention strategies. The methodological approach includes the use of machine learning techniques, such as LGBM and decision trees, combined with advanced methods of causal analysis, such as Double Robust machine learning and Conditional Average Treatment Effects (CATE) modeling. The objective is the development of tools that assist in identifying the determining factors for customer defection, encompassing demographic aspects to the nature of services provided, analyzing variables such as contract type, gender, age, among others. It is expected that the results validate the theories of Wu et al. (2021) on churn prediction and unveil customer profiles with a higher propensity for abandonment, significantly contributing to customer relationship management and offering strategic data for the development of more assertive retention tactics.