Helpful or harmful? A curvilinear perspective on AI agent anthropomorphism in service failure tolerance

Abstract

The anthropomorphic design of AI agents often evokes mixed responses, yet it remains a common strategy to mitigate the negative impact of service failures. This study examines how AI agent anthropomorphism influences customer tolerance for service failures. Through three experiments involving 526 participants in video- and text-based scenarios, the findings indicate a curvilinear relationship between anthropomorphism and failure tolerance, following an inverted U-shape. The study investigates the underlying mechanism driving this effect, with likability serving as a mediator in the curvilinear relationship. Additionally, this study identifies boundary conditions of the effects, where failure severity (1) attenuates the curve relationship between anthropomorphism and failure tolerance and (2) weakens the indirect effect of likability. These findings contribute new factors to cognitive appraisal theory (i.e., likability to personal factor and failure severity to situational factor) and expand uncanny valley theory from emotional to behavioral responses (i.e., tolerance behavior). Practically, this study provides valuable insights for the utilization and deployment of AI agents in customer service.