Customer Service Chatbots: From Automation to Collaboration

Abstract

Customer Service Chatbots: From Automation to Collaboration Customer service chatbots are widely used to answer customer questions, guide service processes, and resolve common issues through natural language interaction. Most operate through rule-based intent recognition, matching customer requests to predefined responses. Yet technical performance alone does not guarantee effective collaboration with the humans they are meant to support. This dissertation examines human–chatbot collaboration from organizational, interactional, conversational, and user perspectives. It shows that successful collaboration depends not only on accurate intent recognition, but also on organizational positioning, conversational practices, handover processes, and the management of user expectations. From an organizational perspective, different stakeholders define chatbot success differently. Managers prioritize efficiency and automation, conversational designers focus on interaction quality and structured handovers, and customer service employees value relief from repetitive tasks and adequate preparatory information. These divergent criteria complicate implementation and require explicit alignment. From a conversational perspective, collaboration is shown to be an ongoing process. Informational requests are generally handled successfully, while transactional requests more often lead to misunderstandings and handovers. Effective chatbots actively monitor understanding, explicitly check interpretations, and support well-prepared transfers to human employees. Handover conversations are not merely technical transitions but communicative processes in which customers adapt their language and interactional behavior. From a user perspective, this dissertation demonstrates that design and interface features primarily shape expectations rather than actual task performance. Human-like cues tend to raise expectations, whereas cues about system limitations lower them. User preference depends less on how human a chatbot appears and more on whether its performance matches the expectations it sets. Across the studies, four conditions for successful human–chatbot collaboration emerge: clear organizational positioning with shared success criteria, active monitoring and repair of misunderstandings, timely and well-structured handovers, and realistic expectation management. Understanding in human–machine interaction is conceptualized not as a one-time achievement, but as a dynamic, interactional accomplishment. In the context of generative AI, these findings underscore that fluency and apparent human-likeness are insufficient foundations for trust. Responsible deployment in customer service requires prioritizing collaboration over autonomy, maintaining transparency about system limitations, and ensuring sustained human involvement when interactions become complex or sensitive.