Measuring service quality based on customer emotion: An explainable AI approach

Abstract

This paper develops an explainable artificial intelligence (AI) approach to measuring service quality in voice-based service encounters. Drawing from the psychology and computer science literature, we construct features of a customer's emotion dynamics during a service encounter. Using real-world call center data from a large insurance company, we train an ensemble model with these emotion dynamics features to predict service quality. The model has higher prediction performance than the two benchmark approaches using quality-assurance evaluation and operational indices. Our method for emotion dynamics classification outperforms a host of state-of-the-art time series classification algorithms. We further apply explainable AI methods to identify the most important features of emotion dynamics and show how they are related to service quality. For example, the location where the last emotion episode appears in a service call has a U-shaped relationship to low quality. Finally, to demonstrate utility, we design an IT artifact to automatically measure service quality after service encounters in the call center and use the measure to predict a customer's referral intention.