The least trustworthy number in most discovery decks is the price. Not the market-size estimate, not the projected retention curve. The price a customer told you they would pay. A 2020 meta-analysis in the Journal of the Academy of Marketing Science pooled 77 studies and more than 24,000 stated willingness-to-pay responses and found that, on average, what people say they will pay runs about 21 percent above what they actually pay once real money is on the line. An earlier meta-analysis of stated-preference studies found the median ratio of hypothetical value to real value was 1.35, meaning the survey number was a third too high before anyone did anything wrong.
That gap has a name in the research literature: hypothetical bias. It is not a sign that your respondents lied or that your survey was sloppy. It is the predictable result of asking someone to price something they are not being asked to buy. And it is the single most common way a product team talks itself into a price the market will not support.
Why the stated price runs high
When you ask “what would you pay for this,” the respondent faces no consequence for answering generously. Saying a high number costs nothing and signals enthusiasm, which most people would rather signal than the opposite. Economists call the low-stakes chatter around this kind of question “cheap talk,” and it inflates estimates in a specific direction every time: upward.
The bias is not uniform. The 2020 meta-analysis found it grows for higher-priced products and for specialty goods, exactly the categories most B2B software and premium products fall into. If you sell a $49 consumer subscription, discount the stated number some. If you sell a $40,000 enterprise contract, discount it hard. The more the purchase would actually hurt, the more freely people overstate what they would tolerate when the pain is imaginary.
This is the same say-do gap that shows up everywhere else in discovery. People are unreliable narrators of their own future behavior, which is why what users say rarely matches what they do. Price is just the place where the gap is most expensive to ignore, because you build an entire business model on top of the number.
The three survey instruments, ranked by how hard they make it to lie
Pricing research is not one method. It is three common ones, and they differ mostly in how much they force the respondent to make a real choice. That framing matters more than the mechanics, because the harder the instrument makes it to answer without a trade-off, the less the answer lies.
Van Westendorp Price Sensitivity Meter. Introduced by the Dutch economist Peter van Westendorp in 1976, it asks four questions: at what price is the product too expensive, expensive but acceptable, a bargain, and so cheap you would doubt its quality. Plot the answers and the intersections define a range of acceptable prices. It is fast and it is popular, and its own critics are blunt about the catch. As one pricing consultancy argues plainly, the method is built entirely on hypothetical answers and it anchors on the point of least customer resistance. It tells you where price stops feeling comfortable. It does not tell you what anyone will buy. Treat its output as a range to investigate, never a price to set.
Gabor-Granger. You show a respondent a series of specific prices and record purchase intent at each, then build a demand curve and a revenue-maximizing point. It is more decision-shaped than Van Westendorp because the person is reacting to real numbers, and Sawtooth Software’s own explainer is candid about its blind spots: it tests your product in isolation, with no competitor on the screen and no anchor, and it still measures stated intent rather than behavior. Directionally useful for tuning a price on a product people already understand. Not a substitute for watching them buy.
Conjoint analysis. Respondents choose among bundles of features and prices, forcing trade-offs the way a real purchase does. Because it makes people give something up to get something else, and because it can put competitors on the same screen, it is the most defensible of the survey methods. It is also the most work, and it is still a survey. The person is choosing, but they are not paying.
Notice the pattern. Van Westendorp asks for a number with no trade-off. Gabor-Granger asks for intent against a number. Conjoint forces a choice between options. Reliability climbs as the exercise moves closer to an actual decision, and it never fully arrives, because none of the three involves money leaving an account.
What I learned watching buyers who had already signed
Twenty years in IT operations at a large enterprise telecom taught me to distrust what buyers say about price, because I was one of them. In vendor selection, we would tell a supplier in the RFP that cost was the deciding factor. It rarely was. Renewals went to the vendor whose tool our teams had already wired into their daily workflow, often at a premium over a cheaper competitor, because ripping it out cost more in disruption than we would ever save on the line item. What we said mattered to us and what actually moved our money were two different things, and the vendors who believed the first one negotiated against a phantom.
The same lesson arrived from the other side of the table in a fractional COO engagement through Ops Harmony. A company had set a subscription price off a Van Westendorp study that put the acceptable range around $49. The survey implied healthy demand there. Trial-to-paid conversion at $49 came in at less than half what that range suggested it should. The survey was not fraudulent. It measured what people said when nothing was at stake, and nothing at stake is not the condition under which anyone buys software. We stopped surveying the price and started testing it.
The number that doesn’t lie is the one attached to a card
Revealed preference is the antidote, and it does not require a working product. It requires a real commitment.
The strongest version is a pre-sale. When Tesla wanted to know whether people would buy a Roadster before one existed, it asked for a $5,000 deposit to hold a build slot. A deposit is a willingness-to-pay measurement with the hypothetical removed. A softer version is the pricing-tier fake door: put the plan or add-on on your live pricing page and count who clicks upgrade or contact sales before the feature behind it exists. The fashion site Polyvore ran an even leaner version of this idea years ago, adding a “buy this outfit” button with no store behind it and fulfilling the first orders by hand; roughly 2 in every 100 people who saw an outfit clicked to buy. That 2 percent is worth more than any survey, because every click was a person acting, not answering.
The company I worked with ran a version of the tier test on its own pricing page. The price that actually converted was lower than the survey’s optimum and structured differently, per seat rather than a flat monthly fee, because the value scaled with team size and buyers priced it that way with their behavior even though they had never said so in an interview. The same logic underlies why an exit survey is the least honest data you own: stated reasons drift toward the socially acceptable, while the action, cancelling, staying, upgrading, is the truth.
A sequence that treats price as a hypothesis
None of this means throw out surveys. It means put them in the right order and stop letting them make the final call.
Use the survey work early and for what it is good at: bounding the range and finding the value metric, the unit your customer mentally attaches cost to (a seat, a transaction, a gigabyte, a location). Van Westendorp will hand you an outer boundary. Conjoint will tell you which features carry the price and how competitors reshape the choice. Treat every figure that comes out as a hypothesis with a known upward bias, and write down the discount you are applying before you see whether you like the answer.
Then validate the specific price with behavior before you commit the roadmap and the financial model to it. A pre-sale, a deposit, a pricing-page test, a cohort you actually charge. You are looking for the point where stated intent survives contact with a payment field, and it will land below where the survey pointed. Pricing analysts who run these studies for a living routinely discount stated numbers by 15 to 30 percent for exactly this reason, which happens to bracket the 21 percent the meta-analysis found. When your behavioral test and your survey range disagree, the behavioral test wins. It is measuring the thing you actually sell.
Whether a problem is worth solving and whether a market will pay to solve it are two separate questions, and the price survey answers neither on its own. It gives you a hypothesis dressed as a finding. The discipline is remembering which one it is, applying the discount before the number flatters you, and refusing to set a price until you have watched at least a few people reach for their wallet.