Willingness to pay is not a survey question

Sample text, to show how a post reads. Replace with your own.

Ask a customer what they would pay for something and you will get an answer. The answer will be delivered with confidence, it will be internally consistent, and it will not predict what they do when the invoice is real.

This is not because people lie. It is because the question asks them to forecast their own behaviour in a situation they are not currently in, with no consequence attached to being wrong. Nobody is good at that.

What goes wrong

Direct questions produce two predictable distortions. Respondents anchor on whatever number appears in the question, or on what they pay today. And they under-report, because saying a lower number costs nothing and feels shrewd.

The result is a figure that looks like data, gets averaged, and ends up in a pricing decision — where it quietly encodes the survey's framing rather than the market's behaviour.

The useful question is never "what would you pay". It is "which of these would you choose", asked enough times, with price varying, until the trade-offs reveal themselves.

What works instead

• Force a choice between configured alternatives at stated prices, rather than asking for a number in the abstract.
• Vary price across respondents so the sensitivity comes from the design, not from the respondent's introspection.
• Validate against transaction data wherever it exists. Observed behaviour outranks any stated preference.
• Look for the shape of the demand curve, not a single threshold. The interesting question is how fast take-up falls, not where it stops.

None of this is exotic, and it is considerably cheaper than discovering the answer after a repricing.