When Service Automation Fails, the Handoff Decides What Happens Next

Customer service technology tends to be judged on the clean moments. The fast answer. The neat routing. The customer who gets what they need without waiting in a queue. Yet the more revealing test comes when the system fails.

A recent study by Chang et al., (2025) looked at what happens when customers experience a service failure while using either an AI agent or a traditional Interactive Voice Response system, better known as IVR. The research focused on product consultation calls at a large Chinese airline, where customers were asking about tickets and luggage services. Some callers were routed through an AI voice agent. Others went through IVR. When the self-service stage failed, customers moved to a human agent for recovery.

The finding is useful for anyone building customer service systems. Customers who experienced a failure with an AI agent were more likely to make a purchase during the human recovery stage than customers who had experienced a failure with IVR. At first glance, that sounds counterintuitive. A failure is still a failure. But the study suggests that the type of failure, and the way the customer has been involved before reaching a person, can shape what happens next.

Why IVR Still Feels Like a Dead End

IVR has been part of customer service for decades, and most customers know the rhythm too well. Press one for this, press two for that, listen to a menu, repeat information, go back to the start and listen properly the second time, wait, wait and, hopefully, all is resolved. It can work for simple routing, but it often asks the customer to fit their problem into the company's structure and increases customer frustration after too many internal transfers.

That makes failure feel especially frustrating. When IVR breaks down, the customer has usually spent time navigating options rather than explaining what they actually need. The interaction is narrow. There is little room for nuance, and very little sense that the system has understood anything beyond a button press or short command.

In the study, IVR failures were compared with AI agent failures because both sit in the self-service stage before a human agent becomes involved. The difference is in the interaction style. AI agents allow customers to express their needs through speech, in a more conversational way. Even when the AI agent does not solve the issue, the customer may have already begun describing the problem, thinking through the purchase, and engaging with the service process.

That extra involvement is important and nuanced.

The Role of Customer Involvement

The researchers found that customer involvement helped explain why AI agent failures led to stronger purchase behaviour after human recovery. Involvement was measured by the number of interaction rounds between the customer and the human agent, with more rounds suggesting deeper engagement.

This makes sense in a service setting. A customer who has already explained their needs through a conversational system may arrive at the human agent with more context, more intent, and more willingness to keep going. They may also feel partly involved in the process rather than completely blocked by it.

The study draws on attribution theory, which looks at how people assign blame when something goes wrong. When customers are more involved in a service process, they may be less likely to place all responsibility for the failure solely on the company. A customer who has interacted with an AI agent, spoken through their request, and then continued with a human agent may experience the failure differently from someone trapped inside a rigid phone menu, experiencing vertigo from too many re-directs and transfers.

That does not mean AI failures are harmless. Poor automation can still damage trust, increase frustration, and push customers away. The point is more specific. A conversational failure can leave the door open for recovery if the human agent picks up the moment well.

The Human Recovery Moment Is the Commercial Moment

The study's strongest practical message is that automation and human support cannot be treated as separate experiences. Customers do not judge the AI agent, the IVR system, and the human agent in neat compartments. They experience one service journey.

For service leaders, that makes the recovery stage commercially important. If a customer reaches a person after an AI failure, the human agent needs the right context, a clear summary of what happened, and enough authority to move the conversation forward. A weak handoff wastes the involvement already created. A strong one can turn a failed self-service interaction into a completed purchase.

This also changes how companies should think about automation performance. The aim should not only be containment or call reduction. Those numbers can hide what happens after a failure. A more useful question is whether the system keeps customers engaged enough for a human agent to recover the interaction properly.

The research does not suggest that every company should rush to replace IVR in every setting. It was based on a specific airline context, and the calls related to product consultation rather than every type of customer service issue. Still, the lesson travels well. Customers are more willing to stay with a service journey when the system lets them participate in it.

AI agents will fail. IVR systems will fail. Human agents will fail too. What separates a damaging failure from a recoverable one is the design of the path that follows. If the customer has been heard, if the human agent can see what happened, and if the conversation continues without forcing the customer back to the beginning, recovery has a real chance.

Source

Li, B., Chang, Y., Liu, L., Liu, H., & Sun, J. (2025). How does AI agent (vs. IVR system) service failure impact customer purchase behavior: Mediating effect of customer involvement. The Service Industries Journal, 45(7–8), 702–720. https://doi.org/10.1080/02642069.2024.2344113