Answering got cheap
Listening didn't

Earlier this year the Capgemini Research Institute asked executives at large companies a simple question: would your customers recommend you? Most said yes. Then it asked the customers. Fewer than half would. 84 against 45, if you want the numbers. I don't think the executives were lying, and I don't think they didn't care. I think they'd stopped hearing their customers, and nothing on their dashboards told them so.

I'd have been in that 84%. The customers I've been most confident about over the years were usually the ones I'd stopped sitting down with. You don't notice the moment it happens. The account looks healthy, the renewal came through, the tickets are closing. Then one day the call comes, and you find out they've been unhappy for a year and you were the last to know.

What it feels like from their side

Think about the last time you had a problem with a company you pay. You explained it to a chatbot. It couldn't help, so you asked for a person. The person asked you to start from the beginning. Then another person did. By the third telling you weren't explaining the problem anymore. You were deciding whether to stay. A week later a survey arrived asking how they did.

That's where most customer relationships actually end. Not in a dramatic failure. In the quiet discovery that nobody was listening. Three in five people in the Capgemini study said they often have to re-explain their problem to a human after a chatbot. Medallia's 2026 report found that in many companies a third or more of departments take no action on the feedback they collect, and that fewer people bother to answer surveys every year. People stop telling you things when telling you hasn't changed anything.

What AI changed, and what it didn't

AI has made answering nearly free. A question gets a reply in seconds, at any hour, in any language. That's genuinely good, and customers are happy to use it for the simple things. But answering and listening are different acts. Answering is about resolving the question in front of you. Listening is about finding out what the customer meant, and what they didn't say, and carrying that forward so the next person doesn't make them start over and so the thing they complained about actually changes.

The reply got cheap. Listening didn't. If anything it got more valuable, because now the moments when a person is in the conversation are the moments the automated reply wasn't enough: the complaint, the escalation, the renewal, the first meeting with someone deciding whether to trust you. The customer is telling you the most in exactly those moments. The person across from them needs to be good at hearing it.

Klarna learned this in public

In early 2024 Klarna announced its AI assistant was doing the work of 700 customer service agents. Fifteen months later its chief executive told Bloomberg the company was hiring people again. The AI had been cheaper, he said, and the quality had been lower, and he wanted customers to know there would always be a human if they wanted one. I'm relaying that from Entrepreneur's report of the interview rather than the interview itself.

What strikes me isn't that the AI failed. It did what it was measured on. The measure was cost. Cost can't hear a customer.

Listening is a skill, and you can see who has it

Listening isn't a personality trait. It's a handful of things a person does, and you can watch for them. They ask the second question instead of jumping to the fix. They say back what they heard and let you correct it. They write it down in your words, not the ticket category. They carry it to the next person so you don't start over. And they come back later and tell you what changed because you spoke up.

Every one of those is a people skill. None of them is the model's job. If I were building a customer-facing team this year, those are the skills I'd hire for and train for, and I'd let the AI handle the rest. The same goes for the people who sell, the people who run accounts, and the people who run the company. Nobody is senior enough to skip it.

The new approach

Pick five customers this quarter and talk to them yourself. Not a survey, not a review deck. Ask what they'd change if they could, and write the answer down in their words. Then put it next to what your dashboard says. If the two agree, good. If they don't, you've just found the gap between 84 and 45 inside your own company, and you found it before the customer did.

Then make sure somebody owns listening the way somebody owns automation. Who reads the conversations. What they're looking for. How what they find reaches the people who can change things. And when something changes because a customer spoke up, tell them. That's the moment they decide you were worth talking to.

The gap in that survey isn't a technology gap. No model closes it, because the model only knows what you already believed. Someone has to go and hear it. That's still a person's job, and I think it's the most important one left.

Sources, all read 16 October 2026: Capgemini Research Institute, "Reimagining customer experience," 2026, a survey of 9,500 consumers and 1,200 executives and front-line staff run in November and December 2025, for the recommendation gap and the re-explaining figure. Medallia, "2026 State of Customer Experience Report," press release, 5 March 2026, for the no-action and survey-response findings. The Klarna account is relayed from Entrepreneur, 9 May 2025, reporting a Bloomberg interview I couldn't read directly. Gartner, the World Economic Forum and ServiceNow all publish related findings that I couldn't verify at the source, so they're left out.