Unraveling the curvilinear impact of AI agent anthropomorphism on service failure tolerance

Anthropomorphic AI agents are increasingly prevalent, yet their impact on customer responses to service failures remains contentious. Drawing on cognitive appraisal theory, this study explores the curvilinear effects of AI agent anthropomorphism on customer service failure tolerance. Through two video-based experiments and one text-based experiment involving a total of 526 subjects, we aim to investigate whether an inverted U-shaped relationship exists between anthropomorphism and failure tolerance and if likability serves as a mediating factor in this relationship. We further explore the potential moderating role of failure severity, hypothesizing that high severity may attenuate the effect of anthropomorphism on likability while low severity could strengthen the indirect relationship between anthropomorphism and failure tolerance via likability. These insights are expected to enrich cognitive appraisal theory and the uncanny valley effect by uncovering potential internal emotional mechanisms, offering valuable implications for designing and implementing AI agents.