ARTICLE · Research

The customer is not lying — they simply do not know why they chose

Boris Kaptelov · 02.08.2026 · 8 min

Nothing in a research report looks more convincing than a customer quote with a conclusion underneath. "I switched to them because our stock counts never matched" — followed by a recommendation to rework the warehouse module.

The trouble is that people often have no direct access to why they chose something. What they have access to is a plausible explanation. These are not the same thing, and the difference between them costs money.

Four different things that look identical in a report

When a customer tells the story of a purchase, four distinct objects are present in the conversation at once. A good report keeps them apart. A bad one fuses them into a single narrative, and opinion then reads as fact.

The first is what actually happened. The second is what the person remembers about it. The third is the explanation they give. The fourth is the explanation the researcher constructs.

They may overlap only in part. The richness of detail, the speaker's confidence, and the emotional force of the story establish, by themselves, neither the accuracy of the memory nor the cause of the action. The smoothness of a story says nothing at all about its accuracy — yet it reads as a mark of credibility.

Memory is not stored — it is reassembled

A common assumption holds that a memory sits somewhere in its entirety and a question simply retrieves it. The Conway and Pleydell-Pearce model describes a different mechanism: an autobiographical memory is a temporary construction, assembled at the moment it is requested, from a general base of self-knowledge and under the influence of the person's current goals.

This has a practical consequence. The same customer, asked under different circumstances, will assemble different versions of the same episode. Not because they are inconsistent, but because the assembly is new every time.

The second consequence matters more for our purposes. When people are asked about a recurring activity, they do not run through the episodes and count them. They take a few accessible instances and fill in the rest from their general notion of the process. So the answer "it usually takes about two hours" may be a memory of one specific case, an average of a couple of memorable ones, or an off-the-cuff estimate. There is no way to tell them apart in conversation — yet decisions get made on those two hours.

Third: memory holds on well to the rare, the vivid, and whatever ended badly. Routine minor costs it holds poorly. An interview corpus is therefore systematically skewed toward failures and peaks and underestimates the steady everyday losses — precisely the ones that usually make up the bulk of the total.

Knowing the outcome rewrites the story of the decision

Hindsight bias was described by Fischhoff as far back as 1975: knowing the actual outcome raises a person's subjective estimate of how predictable that outcome was in advance, and changes which early signals now seem significant to them.

In purchase interviews this cuts both ways. A customer whose implementation went well will tell a coherent story in which everything led to the right choice. A customer whose implementation failed will recall early doubts as obvious warnings they failed to heed.

Both stories are sincere. Neither reproduces the field of alternatives the person actually saw at the moment when they did not yet know how things would end. And that is the field in which the decision was made.

A confident explanation of a choice that never happened

In 1977, Nisbett and Wilson pulled together the experimental evidence showing that people often have no direct access to the processes linking a situation to their response. Asked for a reason, a person does not observe their inner mechanism. They substitute a culturally plausible theory of what usually causes decisions like this.

How far this goes was shown by the experiment of Johansson and colleagues, published in Science in 2005. Participants were asked to choose between two photographs; the chosen one was then covertly swapped for the rejected one, and they were asked to explain their choice. Some participants failed to notice the swap and confidently explained why they had chosen the face they had in fact rejected.

It does not follow that customer explanations are useless. What follows is a precise account of what they are. The answer "I chose them for the quality" reliably tells you that quality is an acceptable reason for this person, that this is the word they use to describe their choice, and that the category matters to them. It does not tell you that quality was the cause. The choice was shaped simultaneously by availability, price, habit, someone else's recommendation, and the way the options were arranged.

What a person protects in a conversation besides the truth

The review by Tourangeau and Yan shows that misreporting on sensitive questions is widespread and depends not only on stable personal traits but on the situation: whether the answer contains potentially embarrassing information, who is asking, and what disclosure might lead to.

In a consumer interview, the socially approved answers are rational spending, care for one's health, and environmental concern. In a corporate interview, add to that competence, compliance with procedure, a successful implementation, and loyalty to a decision the company has already made.

Alvesson describes this for organizational interviews outright: the respondent may be speaking simultaneously as a user, as a representative of their function, as the salesperson of an internal project they pushed through, and as the defender of their company's decision. The gap between account and practice here is not a lie. The person is protecting face, role, employer, and the agreed version of events — and doing so in good faith.

And what makes a question sensitive is not the topic but the position of the particular person. A question about why a procedure was bypassed is neutral for one respondent and threatening for another.

The interviewer is part of the data, not a channel for it

One more assumption worth naming out loud: that knowledge sits ready-made inside the respondent and the question merely transfers it.

The wording of a question frames the answer. If the words "loss," "manual work," or "inefficiency" were first uttered by the interviewer, the respondent will go on describing their work in those categories, even when no one was trying to lead them. A pause, a show of surprise, a nod of approval, and the decision to probe this topic rather than the one next to it also take part in creating the material.

Who is asking changes the answer too. A customer will describe the service differently to a vendor's employee, an independent researcher, and an industry expert. An employee will describe a process differently to their manager, a consultant, and a colleague from another function.

This cannot be eliminated. It can be recorded: who asked, on whose behalf, and in what words the topic was introduced.

What each type of evidence actually confirms

All of the above can be condensed into a single discipline for reading a report.

What to do about it

Three things that cost nothing and change the quality of the conclusions.

Ask about the instance, not the reason. "Tell me how it went last time" yields material that can be cross-checked. "Why did you choose" yields the respondent's ready-made theory. The cause is something the researcher builds by comparing cases, not something copied out of an answer.

Look for what contradicts. One case that does not fit the explanation sharpens the mechanism more than a tenth confirmation of what is already understood. How this turns into a budget line, and why a negative case is not always informative, is covered in our article on budgeting a research project.

Demand that conclusions be checked against traces. A claim that approval takes a week is verified against timestamps in the system. A claim about the cause of churn — against cohorts. When the account and the trace agree, that is a different level of proof, and it is exactly what you should demand of a report, whether your own or a contractor's.

The limitation of the interview is not that people say things that are untrue. It is that the lived event, the accessible memory, today's meaning, the public explanation, the intention, and the future action are six different things. A good researcher does not confuse them. A good client checks that they don't.

If you are holding a report with confident causes of churn or of choice, taking it apart with this framework is a few days' work. It usually changes not the conclusion itself, but which decisions the conclusion can legitimately support.