Self-recovery or human Intervention? understanding the role of task type and failure frequency in chatbot failure recovery

Abstract

Frequent failures of chatbots in customer service highlight the need for effective recovery strategies, such as chatbot self-recovery and human-agent recovery. However, the optimal implementation of these strategies remains unclear. This study, drawing on Task-Individual-Technology Fit (TITF) theory and Mental Accounting Theory (MAT), explores how recovery strategies (chatbot self-recovery vs. human-agent recovery) interact with task types (search task vs. complaint task) to affect recovery satisfaction. It also examines the psychological mechanisms and boundary conditions underlying these effects. We conducted two pilot studies and two online experiments. For search tasks, we found that chatbot self-recovery leads to higher customer satisfaction than human-agent recovery. In contrast, for complaint tasks, human-agent recovery leads to greater satisfaction than chatbot self-recovery. Perceived convenience and perceived empathy mediate these effects. Notably, when chatbots experience double failures, customers consistently prefer human-agent recovery over chatbot self-recovery, regardless of the task type. These findings offer valuable insights for optimizing service recovery strategies and improving customer experiences in hybrid customer service systems.