Seminario interno "Churn Prediction through Customer Feedback Analytics"

18 de agosto, 13:00 horas, sala 504.

Autores: Carolina Martínez, University of Manchester
Babis Theodoulidis, University of Manchester
David Díaz, Universidad de Chile

The aim of this paper is to explore the determinants of switching behaviour embedded into unstructured customer feedback in order to predict churn. Retail banking industry shows a competitive environment with high switching rates despite of it operates under a contractual setting. We used one year direct requests and complaints as a customer feedback data. An integrative approach of frameworks was applied to deploy text mining. The determinants of switching behaviour was extracted through a generic theory developed by Keaveney in 1995 within the activities, resources and contexts involved, i.e. ARC model in order to resolve the uni-linear perspective of Keaveney model. Based on the text mining results, classification algorithms, such as, logistic regressions, decision trees and support vector classifiers are later used to predict churn. The main findings show that the customers tend to articulate determinants of switching behaviour within their requests and complaints. These determinants might be used as predictors of churn to generate actionable information to develop customer engagement strategies in real time. In addition, the extended ARC model provided fruitful insights to comprehend the co-creation of value process in this particular domain.