ISSUE 003
A biased pricing method that predicts well anyway
The last two entries in this pillar were about the limits of asking people things. This one is the other half, because there is a pricing method that is demonstrably biased, widely used, and — when somebody checked it properly — turned out to predict real willingness to pay about as well as a method where participants risked actual money. Understanding why is more useful than the method itself.
An idea worth thinking about
Ask about price directly, or hide it among other things and infer it
Pricing research splits into two families and knowing which you're in prevents most of the confusion. Direct techniques assume people have some sense of what a thing is worth and therefore that asking about price explicitly makes sense — the open question "how much would you pay for this," and the more structured methods built on top of it. Indirect techniques, typically conjoint or discrete choice analysis, bury price among other attributes, ask people to choose between whole packages, and extract the price sensitivity from the pattern of choices.
The trade-off is roughly honesty against realism. Direct methods get you an answer fast and cheap, from someone who knows they are being asked about money and can posture accordingly. Indirect methods are closer to how buying actually feels — you never choose a price, you choose a thing that has a price — and cost considerably more to run.
The caution I'd carry from the consultancies who do this for a living, and which survives their obvious interest in selling the work: pricing research is an input into a pricing decision, not a complete or absolute answer. Anyone offering you a number from a survey and calling it your price is overselling the method.
Read the method comparison
A tool I disappeared into
Four questions you can run yourself this week
The Van Westendorp Price Sensitivity Meter, published in 1976, is four questions about the same product: at what price would it be so cheap you'd doubt the quality, at what price is it a bargain, at what price is it getting expensive, and at what price is it too expensive to consider. Plot the four distributions against each other and the crossing points give you an acceptable range rather than a single number.
It costs nothing but the asking, which is why it's everywhere. The practical guidance from people who run these is to collect at least a hundred completed responses, and to hit that threshold within each segment if you intend to split the results by customer type or plan tier — which is the constraint most small businesses will actually run into.
Know what it doesn't do before you run it. It gives an acceptable range, not a demand curve and not a revenue-optimal price. If you need to model how many people buy at each specific price point, Gabor-Granger asks exactly that and conjoint handles trade-offs. Using Van Westendorp to pick a revenue-maximising price is using the wrong instrument confidently.
See the four questions
Learn something
The gap between what people say they'd pay and what they'd pay has a name and a direction
Hypothetical bias is the term for what happens when willingness-to-pay answers come from a survey scenario rather than an actual purchase. It is not random noise. A study of 215 German consumers pricing mobile phone plans found a moderate negative correlation between a respondent's actual willingness to pay and the size of their hypothetical bias.
Read that carefully, because it's the practically useful bit. The people who would pay the least overstate the most. Which means a survey doesn't just inflate your average price — it disproportionately inflates the answers of the customers least likely to buy, pulling your estimate toward a segment that will not convert. The distortion has a shape, and the shape works against you specifically.
The antidote in the research literature is incentive alignment: designs where the respondent might actually end up buying at the price they state. That is expensive in a lab and nearly free in a business, because you already have the mechanism. Offer the thing at a price and see who pays.
Read the validity research
What we’re seeing
It is biased, and it predicts about as well as making people actually pay
Somebody finally checked the method properly. Kunter compared the Van Westendorp Price Sensitivity Meter against two other approaches: contingent valuation, another direct hypothetical technique, and the Becker-DeGroot-Marschak mechanism, which is incentive-aligned — participants can genuinely end up buying at the price they name, so lying costs them.
Two findings, and they sit oddly together. Van Westendorp does yield biased results, because it is hypothetical and because its construction focuses on the point of minimum customer resistance rather than on what people would actually pay. And its predictive quality is nonetheless high, with results comparable to the incentive-aligned mechanism.
A method can be systematically wrong and still useful if the error is consistent. That is the whole finding, and it is the most practically valuable thing in this entry. It means Van Westendorp is a reasonable instrument for direction — is $40 plausible or absurd, is the acceptable band wide or narrow, does this segment differ from that one — and a poor one for a final number you build a business plan on.
The vintage matters and I'll state it: this comparison was published around 2016. Methods research ages better than market data, but a decade is a decade, and I found no more recent replication.
Where this leaves the three entries in this pillar. Simulated customers fail at predicting individuals. Real customers asked directly are weak predictors of their own behaviour. And a specific, well-constructed, still-biased survey turns out to predict acceptably — but only because someone validated it against a design where people risked money. That is the throughline. Every method in this field is trustworthy exactly to the degree it has been checked against somebody actually paying, and in your own business you can run that check directly rather than reading about it.
Kunter, European Journal of Marketing studies
One question
What's the highest price you've ever tested?
Not modelled — charged, to a real customer, to see what happened. Reply with what you found out.
More in this pillar as it's written.
— Shana
Sources link to the original publisher, never to a summary.