“Have I Answered Your Question Satisfactorily?”: Customer Requests, Intent Recognition Errors, and Repair Strategies in Chatbot Interactions

This study analyzes how request type (informational, transactional), intent recognition errors (misunderstanding, non-understanding), and repair strategies (confirmation, options, rephrase) affect intent recognition in customer service chatbot interactions. We analyzed 200 real conversations with a rule-based, task-oriented chatbot from a Dutch public transport company. Quantitative analyses showed that informational requests were generally recognized and were usually successfully handled by the chatbot, whereas transactional requests were often recognized but were typically redirected to a human agent. Moreover, misunderstandings predominantly occurred with informational requests, while non-understandings were more common with transactional requests. Qualitative analyses showed that repair plays a decisive role in conversational breakdowns. Confirmation fostered mutual understanding by making the chatbot’s intent recognition explicit and instructed customers to clarify requests. Options could help reestablish alignment, but it also constrained the dialogue when predefined choices did not fit customers’ requests. Rephrase often led to repeated misunderstandings. Our study makes concrete how several factors influence successful resolution of requests to customer service chatbots. This is valuable to the practice of interaction design, particularly to improve how smoothly the chatbot can guide the customer during the conversation.