The say-do gap is the distance between what customers tell you they want and what they actually do when it is in front of them. As of August 2026, after fifteen years building platform products in healthcare and clinical education, the sharpest lesson I have on this came from being wrong in public. Customers told us they did not want advertising. I did not want advertising either. The business shipped it anyway, and the usage data disagreed with all of us — not by proving the customers liars, but by revealing that what people say about a hypothetical and what they tolerate in practice are two different measurements of two different things. Both were true. Only one of them paid for the platform.
What is the say-do gap?
The say-do gap is the distance between what people tell you they will do and what they actually do. In product work it is the difference between survey responses, interview answers and support tickets on one side, and usage data on the other. Both are real evidence. They frequently disagree, and the disagreement is the useful part.
The gap is old enough to have a founding study. In 1934, the sociologist Richard LaPiere published Attitudes vs. Actions, reporting on two years spent travelling the United States with a Chinese couple during a period of open anti-Chinese prejudice. They were refused service at one of 251 establishments. Surveyed months afterwards, roughly 92% of those same businesses said they would not serve Chinese guests.
The study has real problems — six months separated the visits from the survey, the people who answered were often not the people who had served them, and LaPiere himself noted his companions were affluent, Westernised and disarming. Worth saying plainly, because an article about weighing evidence should not wave through its own.
But the shape of the finding survived ninety years of methodological argument, and it is the shape that matters: stated attitude and observed action came apart, dramatically, and the people involved were not lying.
Why does customer feedback disagree with customer behaviour?
Because they measure different things. Feedback captures what someone believes about a hypothetical, filtered through how they want to be seen. Behaviour captures what they chose when something was actually in front of them and the cost was real. Neither is dishonest. People genuinely do not know in advance what they will tolerate.
Ask a clinician whether they want advertising in a reference tool they use between patients and you will get a clear answer. It is a sincere answer. It is also an answer about an imagined product, given by someone with no reason to model the economics of the thing they are being asked about, at a moment when the cost of saying no is zero.
Then ship it, and the same person has a different question in front of them: is this still the fastest way to get what I need right now?
Those are not the same question. We kept treating the first answer as a forecast of the second.
What happened when we shipped it anyway
I was against it. So was most of the product organisation, and we had the feedback to back us up. The commercial argument won, as commercial arguments do when the alternative is a platform that does not pay for itself.
What I expected was a visible cost — people leaving, engagement falling, the feedback getting louder and being proven right.
What happened was quieter and more uncomfortable. People stayed. Usage held. The tool was still the fastest way to answer the question they had, and the advertising was a thing they navigated past on the way. The revenue was real, and it funded work that made the product materially better for the same people who had said they did not want it.
I have thought about that for years and I still do not think the customers were wrong. They were answering the question they were asked. We were the ones who mistook a preference for a prediction.
Should you ignore what customers say?
No. Feedback is the best early signal you have about what people value and where the friction is, and it arrives long before usage data exists. The mistake is treating it as a prediction of behaviour rather than as a statement of preference. Preferences are real — they are just not the same thing as choices.
Here is the frame I use now, and the reason I stopped arguing about which source to trust:
| Source | Strong at | Silent on | Fails when |
|---|---|---|---|
| What they say | Why. Where the friction is. What they value. | Whether they will actually change what they do. | Treated as a forecast |
| What they do | Whether. Under real conditions, with real cost. | Why. What it cost them internally. | Treated as approval |
| What the business needs | What the company can survive doing. | What anyone actually wants. | Treated as strategy |
Every one of those is evidence. None of them is an answer. Product judgment is what you do when they point in different directions — which is most of the time, and is precisely the part no framework hands you.
What is the risk of trusting behaviour over feedback?
That tolerance gets mistaken for approval. People staying does not mean they are happy. It can mean the alternative is worse, or that leaving is expensive, or that they have not yet found the thing that makes leaving easy.
This is the part of the story I would emphasise if I were telling it to a room of product people, because it is the part that gets skipped. Behaviour tells you what someone accepted under conditions you created. It does not tell you what they would choose if a better option appeared. And a better option always eventually appears.
So the honest reading of what we learned is not customers say one thing and do another, therefore discount what they say. It is narrower and more useful:
Feedback told us what people would prefer. Behaviour told us what they would tolerate. The distance between those two was our margin — and margin built on tolerance is borrowed, not earned.
That is a genuinely uncomfortable conclusion for a product person, because it means being right about the revenue and still carrying a debt.
How do you decide when the evidence disagrees?
Ask what each source is actually good at measuring, then ask which question you are trying to answer. Feedback is strong on why and weak on whether. Behaviour is strong on whether and silent on why. Commercial reality tells you what the business can survive. Judgment is deciding what to do when those three disagree.
The practical move is small and almost nobody makes it: write down, before you ship, what result would change your mind. Not the success metric — the disconfirming one. If engagement had fallen, how far, for how long, before we pulled it? We did not have that number, which meant that whatever happened next, we were always going to be able to tell ourselves a story where we had been right.
That is the same discipline I have written about for deciding under pressure and for running a bounded AI pilot: name the stop condition while the answer is still unknown, because afterwards it is not a decision any more, it is a narrative.
Twenty years in, the frameworks I trust least are the ones that promise to resolve this for you. Listen to your customers and follow the data are both advice that stops working at precisely the moment you need advice — the moment the sources disagree.
That moment is the job.