When and How Does Generative AI Outperform Human Agents in Service Recovery?

ABSTRACT Exploring the potential of Artificial Intelligence (AI) in service failure and recovery (SFR) remains a growing area of research. However, understanding how generative AI (GenAI) performs in SFR remains unknown. Using a multi‐method approach comprising two preliminary studies (qualitative interviews and a lab experiment) and four online experiments in different live streaming contexts, we investigate the impact of GenAI compared with human agents on consumer repurchase intention, following single and double deviations. Drawing on Cognitive Appraisal Theory (CAT), we show that GenAI, compared with human agents, leads to higher repurchase intention after a single failure. Conversely, human agents generate higher repurchase intention after a double deviation. We identify empathy and compensation as moderating factors in a double deviation context. Specifically, a high level of empathy leads to GenAI outcomes on par with those achieved by human agents, and compensation enhances recovery outcomes for GenAI beyond those of human agents. These effects are mediated by promotion emotion, while rival mechanisms (i.e., frustration and helplessness) are ruled out. These findings offer actionable guidance to live‐streaming platforms, game developers, and online event organizers on using GenAI for single failures and human agents for complex double deviations, enhancing promotion emotion and repurchase intentions.