Start with how humans process, not what they say they want

Building for humans sounds obvious. Of course you’re building for humans, who else would you be building for?

But most products (tech or otherwise) are actually built for an imaginary human. One who reads instructions, makes deliberate choices, and behaves the same way in real life as they do in a user interview.

That person doesn’t exist.

Asking someone what they want is one of the least reliable ways to find out. We think we know what we’ll do but we usually don’t.

Humans are notoriously bad at predicting their own behavior (Wilson & Gilbert, 2003), bad at reconstructing why they made past decisions (Nisbett & Wilson, 1977), and surprisingly good at telling you what they think you want to hear (Crowne & Marlowe, 1960; Nederhof, 1985). Yet most product research is built on this anyway.

The core problem with most product building is that it relies on some degree of self-report, like surveys, interviews, or stated preferences.

Here’s what researchers say you should actually do.

Instead of asking “Would you use this?” you watch whether people actually use it. The difference is in the result. Asking results in an opinion. Watching, on the other hand, results in evidence. In practice, this means:

A survey asks people to reflect on their experience and report it back to you. The problem is that reflection is reconstructive. People don’t replay what happened but instead build a story about what happened, shaped by how they feel right now and what they think you want to hear (Nisbett & Wilson, 1977).

Analytics skips the reconstruction entirely. Here are popular analytics and what they can do for you:

Clickmaps show you where people actually went.

Session recordings show you where people got confused, hesitated, or gave up.

Drop-off points show you exactly where your product lost someone, without asking them to explain it.

Time-on-task tells you whether something was easy or hard without requiring anyone to say so.

Asking people which version they prefer produces preference data. Showing half your users one version and half another and measuring what they do produces behavioral data. These are not the same thing.

People regularly prefer one version in a survey and behave differently when it’s in front of them for real.

A/B testing removes the opinion entirely and replaces it with a controlled observation.

The version that produces the behavior you want wins, regardless of which one people said they liked.

The most reliable data you can get from a potential user isn’t what they say; it’s what they do when something is actually at stake. Behavioral economists call this revealed preference: real choices, made under real conditions, tell you more about what someone actually wants than any answer they give to a question (Samuelson, 1938).

The cost of the action is what makes it meaningful. In a user interview, saying “yes, I’d use that” costs nothing. It’s socially easy, makes the conversation pleasant, and requires no follow-through.

But asking someone to give you their email address before the product exists is an entirely different ask. It costs something small but tangible: attention, inbox space, a moment of decision-making to oblige, or not.

Low-stakes questions produce low-stakes answers (Ariely et al., 2003). As the cost of the action goes up, the gap between what people say and what they want begins to narrow.

This is why waitlists and pre-orders work as research instruments, not just marketing ones (Ries, 2011).

A list of a hundred people who gave you their email before you built anything is meaningfully different from a hundred people who said they’d use it if you built it.

One is behavioral data.

The other is a compliment.

The core idea is that memory is reconstructive (Bartlett, 1932). When you ask someone to reflect on an experience after the fact, you’re not getting a recording of what happened but you’re getting a story they built about what happened, shaped by how it ended, how they’re feeling right now, and what seems valid in hindsight.

Daniel Kahneman, a psychologist and Nobel laureate whose work on human judgment and decision-making is among the most cited in behavioral science, found that people’s overall assessment of an experience is disproportionately shaped by 1) its most intense moment and 2) its final moment, not by an accurate average of the whole thing (Kahneman et al., 1993).

He called this the peak-end rule.

Ask someone on Friday how their week went, and you’ll get a narrative.

Ask them Monday through Friday as it’s happening, and you’ll get a more data-driven response.

So you catch people during the experience instead of after it. In practice, this means:

A one-or two-question prompt sent immediately after a specific action: “You just finished a task. How clear were the instructions?” Captures the experience while it’s still live.

The same question asked in a follow-up email three days later captures a memory of the experience, which is different and less reliable. The timing is a methodological one rather than a convenience (Csikszentmihalyi & Larson, 1987).

A micro-survey that fires immediately after someone completes onboarding, hits an error, or abandons a flow is catching them at the moment of highest relevance.

A survey sent as a follow-up email asks people to reconstruct a moment that has already been filtered through everything that has happened since.

The through-line across both: you’re replacing “what do you think” with “show me what you do.” The former is easy to collect and mostly wrong. The latter is harder to collect and tells you something.

Ariely, D., Loewenstein, G., & Prelec, D. (2003). Coherent arbitrariness: Stable demand curves without stable preferences. Quarterly Journal of Economics, 118(1), 73–105.

Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge University Press.

Crowne, D. P., & Marlowe, D. (1960). A new scale of social desirability independent of psychopathology. Journal of Consulting Psychology, 24(4), 349–353.

Csikszentmihalyi, M., & Larson, R. (1987). Validity and reliability of the experience-sampling method. Journal of Nervous and Mental Disease, 175(9), 526–536.

Kahneman, D., Fredrickson, B. L., Schreiber, C. A., & Redelmeier, D. A. (1993). When more pain is preferred to less: Adding a better end. Psychological Science, 4(6), 401–405.

Nederhof, A. J. (1985). Methods of coping with social desirability bias: A review. European Journal of Social Psychology, 15(3), 263–280.

Nisbett, R. E., & Wilson, T. D. (1977). Telling more than we can know: Verbal reports on mental processes. Psychological Review, 84(3), 231–259.

Ries, E. (2011). The lean startup: How today’s entrepreneurs use continuous innovation to create radically successful businesses. Crown Business.

Samuelson, P. A. (1938). A note on the pure theory of consumer’s behaviour. Economica, 5(17), 61–71.

Wilson, T. D., & Gilbert, D. T. (2003). Affective forecasting. Advances in Experimental Social Psychology, 35, 345–411.