Why asking customers what they want can give you worse answers

Updated Steve New

In customer research, I have repeatedly seen people tell us they want one thing and then behave differently once there is something concrete to do. We have asked which topic or session people prefer, then watched later clicks or viewing go another way. People have asked for recordings or more flexible access and then barely used them. Highly enthusiastic attendees have sometimes shown less buying intent than quieter ones.

It is tempting to conclude that customers do not know what they want and that behaviour gives us the real answer. I do not find that very useful. Behaviour is noisy too, and a customer's account can be accurate even when their explanation or proposed solution is incomplete.

When the two disagree, the question I find most useful is: what would make both observations true?

Keep the customer's experience, explanation and prescription separate

A customer can usually tell you a great deal about the situation they were in. They know what felt difficult, what they were comparing, what worried them, what they tried and where they got stuck. That is often exactly the evidence the business needs.

What I try to avoid is collapsing three different things into one.

Someone might say that a professional programme was difficult to fit around work. That is an account of their experience. They might explain it by saying there were too many live sessions. That is already an interpretation. If they then say the programme should become self-paced, that is a proposed solution.

I want all three, but I do not treat them as the same kind of evidence. The account of what happened can be a strong clue even when the explanation is only part of the story and the suggested fix would create a worse product.

I have seen versions of this with price, format, session times, replays and requests for more content or support. A real frustration does not automatically identify the right intervention.

Behaviour is useful because reality has constraints

If I am trying to understand whether somebody is likely to spend money or change behaviour, I usually want the recent history: what they tried, what they paid for, what they attended, what they postponed and what they chose when money, time or inconvenience were real.

That usually tells me more than asking whether an idea sounds useful, because the earlier decision had to survive real constraints.

But behaviour is not self-explanatory. A click can come from a stronger title rather than deeper interest. Attendance can reflect timing. Email data has bots and other noise. A replay request followed by very little replay use does not prove the request was dishonest; the person may have wanted reassurance that they would not miss out, rather than genuinely expecting to watch hours of material later.

I use behaviour to test and sharpen the explanation, not to overrule the customer by default.

Contradictions are evidence

We have run versions of topic polls where stated preference pointed one way and later clicks or viewing another. Before inventing psychology, I check the boring things: were these actually the same people, seeing comparable choices, at roughly the same point in the decision?

If that still leaves a conflict, there are several plausible explanations. One title may be stronger. People may say what they think they ought to learn while clicking the problem that feels urgent now. Timing or framing may have changed. A vocal subgroup may simply not represent the wider audience.

At that point the original answer can no longer carry the diagnosis by itself. The same thing happens when people ask for a replay and barely use it, or when a few forceful requests make a change sound universal and broader participation data says otherwise. The request may be perfectly sincere and still tell us more about the frustration than the right fix.

What I do when the signals disagree

I try to preserve the observations before turning them into a story. I write down what the customer actually said and the relevant behaviour separately: purchase, click, attendance, reply, replay use, previous spending, support history or whatever matters for the decision.

Then I check whether the comparison is fair: the same people, the same time period, the same offer and framing, and no obvious changed constraint. If not, I may not have found a paradox at all; I may just be comparing different situations.

If the conflict survives that check, I look for explanations that can accommodate both observations and ask what further evidence would distinguish them. What I want is an explanation that survives the next piece of evidence.

It also stops customer research becoming a search for quotes that support a decision we already want to make. I want the awkward evidence left in long enough to learn from it.

When the evidence lines up

When customer language, transactions, support patterns and observed behaviour keep pointing the same way, I am usually happier to act. There is no prize for another research round after the decision is clear enough.

A genuinely new product, a first-time purchase or a major change in circumstances may not have a useful behavioural history. In those cases good conversations have to carry more of the evidential load, so I pay closer attention to what that source can and cannot establish.

Customers are often excellent witnesses to their own situation. I want to know what changed, what they compared, what felt risky, what they had already tried and what happened next. I get more cautious as the answer moves from describing the experience to explaining the cause, and more cautious again when it becomes a prescription for what the business should build.

I keep the words, behaviour and context separate until they make sense together. Agreement usually means I can act with more confidence. Disagreement is a reason to keep looking, not to tidy one source away because it makes the research easier to summarise. Customers can be right about the problem and wrong about the fix. So can I.