Ulrich Gnewuch, Fabian Reinkemeier
5 min
As organizations increasingly rely on AI-powered chatbots for customer service, conversational breakdowns—situations where the chatbot fails to understand a user's request—have become a major source of customer frustration. Current industry standards typically handle these failures in one of two ways: either by forcing the customer to rephrase their input (customer-driven) or by having the chatbot guess the user's intent without acknowledging the error (chatbot-driven). Both approaches often turn interactions into a repetitive, unsuccessful guessing game that damages the relationship between the customer and the brand.
Drawing on the Theory of Least Collaborative Effort (TLCE), the authors propose a shift from confrontation to collaboration. They developed a novel design framework, termed "TLCE-bot," which treats the customer and the chatbot as partners working together to resolve communication gaps. Instead of simply asking for a repeat, the collaborative strategy involves the chatbot acknowledging the misunderstanding, providing context-aware suggestions, and inviting the user to participate in clarifying the intent. This design was instantiated in the chatbot of a large international insurance company.
To evaluate the effectiveness of this approach, the researchers conducted a randomized field experiment with 1,352 real-world customer interactions. The results demonstrate that the collaborative repair strategy is superior to existing methods. It not only increases the likelihood that a breakdown will be successfully resolved but also significantly reduces the negative emotional and behavioral consequences of these service failures. By framing the interaction as a joint effort, the chatbot effectively lowers the barrier for users to continue the conversation rather than abandoning the service.
This research provides a critical shift in how we think about human-AI interaction. By moving away from the assumption that chatbots must be perfect or that customers must bear the burden of correction, the study offers actionable design prescriptions for developers. It suggests that the future of effective AI service lies in designing systems that can gracefully handle ambiguity through partnership, ultimately improving both operational efficiency and customer experience.
When chatbots are deployed to automate customer service, it is nearly inevitable that situations will arise in which they struggle to understand customer requests. Unfortunately, the onus of resolving such conversational breakdowns tends to fall on either the customer or the chatbot alone, turning customer-chatbot interaction into a frustrating and often unsuccessful guessing game. Despite indications that customers would be open to collaboration, we know little about repair strategies that involve the customer and chatbot working together to resolve breakdowns. Our research addresses this gap by investigating the design and impact of collaborative repair strategies in customer-chatbot interaction. Drawing upon an integration of the theory of least collaborative effort with research on human-machine communication and customer service chatbots, we propose a novel repair strategy design; we instantiated it in the chatbot of a large insurance company and conducted a naturalistic summative evaluation through a randomized field experiment. Overall, our results suggest that a collaborative repair strategy can lead to more breakdowns being resolved and mitigate the negative impacts of breakdowns on key customer outcomes. Our research offers a new way of thinking about customer-AI service interactions by shifting the narrative from confrontation to collaboration, extends the theory of least collaborative effort by integrating the perspective of customer-chatbot interaction, and provides in-depth insights into breakdown and repair in real-world conversations between customers and chatbots.
Sam: Exactly. And they tested whether that actually makes a difference. They ran a real-world experiment at a large insurance company with over a thousand customers. One group got the old generic error messages. The other got the new, collaborative repair messages. They tracked whether problems were actually resolved, and whether customers abandoned the chat after a single misunderstanding.
Alex: The results were meaningful?
Sam: They were. The collaborative approach led to nearly an eighteen percent increase in successful resolutions. It also significantly reduced the number of people who quit the chat right after hitting a wall.
Alex: That's a notable shift. Though I imagine it didn't work equally well for every type of problem.
Sam: You're right to push on that. It worked well for the most common failure types—like when a user was too brief, or used unusual phrasing the bot hadn't seen before. But it struggled with what the researchers call "cryptic" breakdowns, where the input was essentially meaningless: random characters, or a language the system couldn't process at all. In those cases, the collaborative strategy didn't help much. And that makes sense—you can't negotiate a repair if there's nothing to work with.
Alex: So the bot needs at least a baseline of sense in the message before the partnership can begin.
Sam: That's a fair way to put it. The study is honest about this. It's not a solution to every failure mode—it's a meaningful improvement for the most common ones. The authors also acknowledge that the experiment was conducted within a single insurance company, so the specific categories of breakdowns they identified might look different in other industries. They suggest future work should test these ideas more broadly, including with more advanced AI systems that generate their own responses rather than selecting from pre-written ones.
Alex: So the principle is solid, but the specific implementation is a starting point, not a finished product.
Sam: That's the right framing. The real contribution here is the shift in perspective. It offers a blueprint for designing systems that don't just "fail" when they get confused, but actively help the user find a way forward. The responsibility for making a conversation work stops being placed entirely on the person typing.
Alex: It's a good reminder that the language a system uses when things go wrong matters just as much as the language it uses when things go right. Thanks for walking us through it, Sam, and thanks to everyone listening to ResearchPod.