Microsoft Copilot/Unsplash
Companies are betting that AI chatbots will deliver faster and cheaper customer service. But if you’ve ever tried to circumvent a chatbot and get to a human, you’re not alone. Here’s what three recent studies discovered about when customers will and won’t let AI do a human’s job.
1. Customers avoid chatbots for two compounding reasons. In a study simulating a customer service scenario, participants repeatedly chose between two unlabeled options: One required waiting in line before their request was resolved with certainty; the other skipped the line, but occasionally failed, routing the customer into the line for the first option. Researchers designed the choices so that a person optimizing for time saved should have picked each option about equally often. Instead, participants chose the no-queue option just 28% of the time — a reluctance researchers call gatekeeper aversion, driven by its uncertainty and multistage structure, regardless of who or what runs it. When that same no-queue option was presented as a chatbot rather than a person, adoption fell by another 10 to 20 percentage points — a separate effect called algorithm aversion. Follow-up experiments hint at possible remedies: Offering transparency about what the chatbot can and can’t do, and showing customers the expected wait time for each option, appears to increase chatbot uptake.
2. AI is a better messenger for bad news; humans, for good news. Across several experiments, customers who received a worse-than-expected offer (say, a low resale price) were more likely to accept it from an AI than from a human agent. In one study, 78.6% accepted an AI’s offer, versus 60.4% for a human’s offer. But in another study, when the offer was better than expected, the human agent’s offer was accepted 89% of the time, versus 76% for the AI. The reason: People don’t ascribe human intentions to AI, so don’t regard it as “selfish” when it lowballs them, nor as “generous” when it overdelivers. The effect is strongest when the AI is presented as machinelike; a humanlike persona erodes the advantage for delivering bad news.
3. A simple two-question test can predict whether customers will embrace or reject an AI. A meta-analysis of 163 studies involving over 82,000 participants found that customer preference for AI over humans comes down to two factors: whether the AI is seen as more capable at the task than a person, and whether the task is seen as requiring personalization. When AI is seen as more capable and personalization is seen as unnecessary, such as when forecasting sales or playing chess, people prefer it. In every other combination, people favor humans out of a desire for individualized treatment. Before automating a customer-facing role, leaders should weigh AI’s capability against customers’ expectations of personalized service.