Â鶹´«Ã½

Abstract

News — Since pure voice-to-voice communications mainly characterize the call center context, vocal cues provide a novel lens to comprehend consumer-agent dynamics beyond mere words. This study proposes an analytical framework exploiting speech recognition and interpretable machine learning to convert unstructured audio data into quantifiable measures and examines the impact of agents’ voices in a natural setting. The results show that incorporating agents’ vocal cues into consumer dissatisfaction and callback analysis improves out-of-sample forecast accuracy, with an average improvement of 11.65% and 4.30%, respectively. Vocal cues surpass verbal and demographic variables in predictive importance. An affirmative tone and a relatively quick speech rate are identified as key factors that significantly reduce dissatisfaction and callbacks. Our proposed voice feature framework enhances telephone-based service quality assessment, offers practical insights for agent training, and provides novel insights to improve consumer service operations, ultimately leading to the maximization of financial benefits.