The Customer's First Answer Is a Solution. So Is ChatGPT's.

Ask a customer what they need from a mattress and there's a good chance they'll tell you they want a gel cooling layer. That isn't what they need. It's a solution somebody else already built, and the customer is handing it back to you because it's the only vocabulary they have for the problem. What they need is to stay cool while they sleep, and there are a lot of ways to give someone that besides gel.

That example came from Kristyn Corrigan, a principal at Applied Marketing Science, who gave the very last talk of the PDMA summit in Atlanta in October. She's been doing voice-of-the-customer research for 21 years, and she opened with the idea she said organizations get wrong all the time: "the customer need and the customer solution are two entirely different things." Teams go to customers, ask what they need, get back a solution or a generality, and build it. It's rarely creative, because the customer has only told you what already exists.

Then, in the second half of her talk, she turned the same lens on the AI tools a lot of us now use to read customer feedback for us, and the mistake turns out to be built into them too.

The first answer is almost always a solution

Her tire example makes the point even more clearly. A customer says, "I want a tire with a deeper tread." It's easy to hear that and call manufacturing about a new tread design. But when you ask why they want it, the need underneath is something like, "I need to feel in control when I'm driving in heavy rain," and that opens up a much bigger set of answers: stopping confidently on wet roads, keeping grip as the tires wear, knowing when the tires need replacing.

The fix she described is old and simple. When a customer gives you a solution, ask why. "I understand you're asking for a gel cooling layer. Why is that? What would that do for you?"

She also talked about the gap between what customers say and what they do. Years ago she was interviewing the operator of a compact track loader on a work site, a die-hard Caterpillar guy in branded gear, and she couldn't get him to say a single thing was wrong with the machine. Then they walked over to it, and she noticed a huge gash on the right-hand side. When she asked about it, he told her that anytime he made a tight turn the visibility on the right side was terrible, so he always caught the side. He hadn't been hiding it. He just didn't think of it until he was standing next to the evidence.

That's why the classic voice-of-the-customer research leans on experiential interviews and observation rather than asking people to list their needs. Abbie Griffin and John Hauser's work at MIT found that one-on-one interviews with about 20 customers in a segment surface more than 90% of the needs that interviewing 30 would. You don't need hundreds of conversations. You need a couple dozen good ones, where someone keeps asking why.

Your sales team is not your customer

Right before Corrigan, Chad McAllister of Product Mastery Now ran through the weaknesses he sees most often when he works with product teams, and one of them goes straight at this problem. "Sales gets used as a surrogate for the customer," he said. Sales is close to revenue, which matters, but they have a different objective than understanding what customers need. And the direct contact that would fix it keeps shrinking. "Most metrics say we want a product manager spending about 30% of their time with customers," he said. "And a lot of the product managers I talk to, it's been nothing in the last six months."

When nobody on the team is talking to customers, the gap gets filled with assumptions, and the assumptions usually look a lot like the team. I watched this happen on a product built for high schoolers. The team had "the high schooler" pinned down as a single, tidy persona, and when I asked what research it was based on, the answer was none. Nobody had talked to students. When I did the research, the one high schooler turned out to be at least three genuinely different users, and the biggest, most underserved group was the one that looked least like the people building the software.

That's the most common bias I see on product teams. It isn't one of the exotic ones with a clever name. It's the everyday assumption that you already understand your user, and it slips in exactly where you feel most sure.

AI reads the customer the way the customer talks

There's a reasonable hope that AI fixes this. Customer input is everywhere now – support tickets, sales calls, recorded meetings, Reddit threads, YouTube comments – and Corrigan cited an estimate of more than 265 billion customer service calls a year. No team can read all of that, but a model can.

The catch is that general-purpose AI makes the same mistake the customer does. Corrigan said that even after years of research, AI "does default to that technical spec into the solution. It's just easier to find in conversation." Her firm tested this directly. In a blind test with BEHR Paint, professional analysts, standard chatbots and a model AMS built for this work all read the same customer comments, and the standard chatbots "only focused on the suggested solution and did not accurately identify the core customer need." (AMS sells that specialized model, so weigh the comparison accordingly. The finding about off-the-shelf chatbots holds either way.)

So if you've pasted a pile of support tickets into ChatGPT and asked what customers want, you've probably gotten back a tidy list of features. It reads like insight, and it's really the customers' own solutions, sorted and summarized. Corrigan's warning was about what gets lost when AI gives you the gist: "which is where innovation often comes from." The nuance, the contradiction, the operator who only remembers the visibility problem when he's standing next to the gash.

Ask why, then go watch

None of this means AI is useless for customer research. Corrigan described using it to process the existing feedback first, so a team shows up to customer interviews already knowing what's been said and focused on the gaps, and she said that's compressed some of her clients' research from months to weeks. That seems right to me. The volume problem is real, and AI is good at volume.

But a few things still have to come from people. Somebody on the team has to be able to tell a need from a solution, a target value or an opinion, so they can look at an AI-generated list and say, "that's not a need, that's a gel cooling layer." Every insight should trace back to a customer's actual words, not to a model's paraphrase of them. And someone still has to go stand next to the machine, because the say-do gap doesn't show up in a transcript.

If you only take one habit from this, make it the why. When a customer, a sales rep or a chatbot hands you a feature, ask what it would do for the person who wants it, and keep asking until you get to something that isn't a feature at all.

As Corrigan put it at the end of her talk, "The customer's first answer is rarely the innovation opportunity." The AI's first answer usually isn't either.