AI Chatbots: Service Failure and Recovery Strategy

This study explores customer reactions to service failures from high vs. low anthropomorphic AI chatbots in online retail settings. Customers were more aggressive towards highly anthropomorphic chatbots when the failure occurred during the service process (e.g., inappropriate response) compared to low anthropomorphic ones. When the failure was in the outcome (e.g., payment failure), customer aggression was high for both chatbot types. This effect was fully mediated by the disconfirmation of anticipated expectations.Customers have higher expectations for effective help from human-like chatbots. Consequently, when these expectations are not met during the service process, it leads to a greater feeling of disconfirmation and increased aggression. In contrast, customers were less aggressive towards low anthropomorphic chatbots experiencing service process failures. When the failure was in the outcome, customer aggression remained high for both chatbot types.An apology from a high anthropomorphic chatbot reduced the disconfirmation of anticipated expectations and aggression in process (vs. outcome) failures. However, acknowledging the AI nature of the agent had no differential impact on customer reactions. Interestingly, apologies or acknowledgements from non-anthropomorphic chatbots did not decrease customer aggression. These findings inform strategies for effective chatbot design, ultimately enhancing customer experience.