Every office has that one colleague. The report is late, and it was the printer. The meeting was missed, and it was the calendar app. The client email went to the wrong person, and it was, naturally, autocorrect. Humans have been blaming machines for as long as there have been machines, and probably blamed the weather before that. But what happens when the machine sitting next to you at work is an AI chatbot that talks to the same customers you do? It turns out you now have the perfect scapegoat: a colleague who never argues back.
A research team from China, asked what happens to customers, and to the company, when human customer service agents shift the blame for problems onto the AI chatbot working alongside them. Their results are extremely interesting.
The rise of the human-plus-bot help desk
If you have chatted with an online store recently, there is a good chance you spoke to a bot first and a human later, or perhaps both at once without noticing. Companies love chatbots because they answer routine questions instantly, never sleep, and never ask for a raise. But chatbots still stumble when a customer is angry, confused, or dealing with something unusual. So most firms now run what researchers call hybrid service systems, where AI and people work together. According to industry figures quoted in the paper, about 80 percent of enterprises now mix AI into their customer service.
The authors focus on a particular kind of hybrid setup where the human and the AI act as equal partners rather than boss and assistant. In the company they studied, a leading Chinese maker of smart bathroom fittings selling through a marketplace, the chatbot steps in automatically if a human agent has not replied within two seconds, and keeps chatting until the human returns. The bot and the person are effectively tag-teaming the same conversation.
That arrangement is efficient, but it also creates a tempting opportunity. When something goes wrong, say the customer was given the wrong delivery date, the human agent can simply say, “I’m so sorry, that was the automated system’s mistake.” The bot cannot object. It has no union, no lawyer, and no hurt feelings. It just sits there, silently taking the fall.
Good for the agent, bad for the brand?
What makes this study clever is that it does not assume blame-shifting is simply good or simply bad. Instead, the authors suggest it may help one party while hurting another.
Consider the customer’s point of view. When something goes wrong, the customer mostly wants two things: to understand what happened and to get it fixed. If the human agent says the bot made an error and then carries on helping, the customer gets an explanation and a friendly person solving the problem. Drawing on a psychological idea called cognitive appraisal theory, which says our emotions depend on how we interpret a situation, the authors predict that customers will actually feel better about the human agent in these moments. The agent looks like the hero rescuing them from the clumsy robot.
But zoom out and the picture changes. When a customer hears that the company’s technology messed up, they may conclude that the company itself is unreliable. Tech errors feel systematic and repeatable. If the bot got it wrong once, it will probably get it wrong again. Customers also tend to hold firms responsible for the tools they choose to deploy and believe such failures could have been prevented with better systems or oversight. So the authors expect that while the customer may warmly rate the agent, they may also become less likely to buy from the company again. The agent wins the battle; the brand loses the war.
Following the blame trail through real chats
To test these ideas, the team obtained a rich dataset from the company covering three months, including chat transcripts, purchase records, and agent performance data. They focused on conversations where something had gone wrong, since that is when blame gets passed around. Identifying blame-shifting in thousands of chats is no small feat, so the researchers used a large language model to read the conversations and flag moments when a human agent pointed the finger at the chatbot. To make sure the AI detective was doing its job, three postgraduate students manually checked 200 random sessions, and the AI’s labels matched human judgement almost perfectly. There is a delightful irony here: an AI being used to catch humans blaming an AI.
They then measured two outcomes. The first was how the customer felt, captured by analysing the emotional tone of the customer’s messages. The second was whether the customer made a purchase within two days of the chat. A tricky problem remained. Perhaps the kind of agents who blame bots are also defensive or abrupt in other ways, and it is their overall style, not the blaming, that affects customers. To untangle this, the authors used a statistical technique that exploits short-term ups and downs in each agent’s blaming habit, such as an agent who had a rough couple of days and became temporarily more likely to blame the system. This helps isolate the effect of blaming itself from the agent’s general personality.
Lessons for managers (and the rest of us)
Even at this stage, the study raises a warning that managers should take seriously. Organisations often judge customer service agents by satisfaction ratings. If blaming the bot boosts an agent’s personal score while quietly eroding customer loyalty to the brand, then the incentive system is rewarding exactly the wrong thing. The agent looks like a star in the dashboard while sales slowly leak away.
The fix is not to ban honesty. Sometimes the bot really did make the mistake. But companies may want to train staff to explain errors in ways that reassure customers the problem is being fixed, rather than implying the company’s technology is unreliable. They may also want to track whether agents are reflexively reaching for the bot as an excuse. And they should make sure that any real AI errors are fixed promptly, so that explanations are paired with genuine corrective action.
There is a broader human lesson as well. As AI becomes a colleague in offices, hospitals, banks, and classrooms, it will increasingly be the easiest target when things go wrong. Accountability, however, cannot be delegated to a piece of software. Customers, it seems, know this instinctively. When the bot gets blamed, they do not just blame the bot. They blame the people who put it there. So the next time a customer service agent tells you, “It’s the system’s fault,” you might smile politely. And then perhaps wonder who chose the system.
