Modelo de segmentación de cartera basado en el comportamiento de clientes para la gestión de cobranza

Abstract

Credit sales with direct financing is an important component in the retail sector development. In addition to the sector's own credit risk, the presence of COVID-19 has resulted in an increase in delinquency rates due to a change in customer payment behavior as a consequence of the country's economic situation. An ill-defined collection management can lead to taking wrong and costly strategies, wasting resources and distorting it. This research proposes a portfolio segmentation model based on customer behavior to improve collection management efficiency. Adopting this approach is key to generating management policies and establishing specific collection strategies. The model contemplates the most representative static variables of the client's profile from the bureau, and internal dynamic behavioral attributes. A non-hierarchical clustering approach is considered, with K-means being the methodology to be used. The model is designed in Azure Machine Learning and concluded in JMP for the graphical understanding it provides. The results show a different solution to the traditional ordering, since the classification of clients is commonly based on the degree of portfolio aging. In addition, since distance is the discriminant in the formation of clusters, this may be the element that facilitates ordering, since it prioritizes belonging to one or another cluster; therefore, it is feasible to use this concept as a priority approach for collection management.