Willingness to Pay Research

The methods everyone cites were built for consumer research with hundreds of respondents. You have a list of 900 customers, of whom maybe 60 will answer a survey, split across three segments that behave completely differently.

That constraint is not a reason to skip the research. It is a reason to pick the right method and to be honest about the error bars.

Which method should you use?

Four options, and they answer different questions. Picking by familiarity rather than by question is the most common mistake in willingness to pay work.

Method	Answers	Minimum sample	Effort	Use when
Van Westendorp	What price range is credible	30 to 40	Low, 4 questions	You have no price yet or are repositioning
Gabor Granger	What single price maximises revenue	30 to 50	Low, 4 to 5 steps	You know the range and need a number
Conjoint or MaxDiff	What goes in each package	150 plus	High, needs design and analysis	You are restructuring tiers and have reach
Pricing interviews	Why they would pay, and what for	12 to 20	High per response	Under 60 possible responses, or a new category

Van Westendorp gives you a corridor. Its weakness is that it measures acceptability, not purchase, so the range it produces is usually wider and more generous than reality. Use it to rule out prices, not to pick one.

Gabor Granger gives you a demand curve and therefore a revenue estimate at each price, which is what you actually need for a pricing decision. Its weakness is anchoring: the first price you show shapes every subsequent answer, so randomise the starting point across respondents.

Conjoint is the right tool for packaging questions and the wrong tool for most B2B SaaS teams, because 150 responses in one segment is genuinely hard to recruit. If you cannot get there, MaxDiff on features is a lighter alternative that tells you relative importance without the full utility model.

Below about 60 possible respondents, stop pretending. Run twenty structured interviews instead and treat the output as qualitative. That is not a consolation prize: interviews tell you why a price feels wrong, which a survey never will.

Who should you sample, and what bias does each group bring?

Three populations, three different distortions. Use all three and compare, because each one is wrong in a direction you can predict.

Current customers answer near what they already pay. They passed your price filter, so by construction they are the people who found it acceptable. Surveying only them produces a ceiling estimate that is systematically too low, and it is the most common sampling error in SaaS pricing research.

Churned accounts tell you about the downside edge. Some left on price, most did not, so ask both why they left and what price would have kept them. Their answers skew low and bitter, which is useful as a floor check.

Lost deals are the most valuable and hardest group. People who evaluated you, considered paying, and chose otherwise. They know your product well enough to have an opinion and they had no sunk cost. Recruiting them takes a personal ask from whoever ran the deal, and a small incentive.

Aim for a rough split of 40 percent current customers, 20 percent churned, 40 percent lost deals and prospects. Segment tightly. A single price study that mixes a 15 person startup with a 4,000 person enterprise produces a bimodal distribution and an average that describes nobody.

How do you word the questions without anchoring the answer?

Carefully, and in this order: value first, price second, never the reverse. Price anchoring is the quiet failure mode that invalidates most in house pricing surveys.

Start by describing the product in outcome terms, not feature terms, and keep the description identical for every respondent. Mention no current price, no competitor price and no range. If your respondent knows you charge 200 a month, their answer will orbit 200 regardless of what they truly value.

For Van Westendorp, the canonical four questions work but the unit matters enormously in B2B. Ask per user per month, per year, or total contract value, and be explicit which one. Respondents who answer in different units will quietly corrupt the dataset and you will not notice until the curves look strange.

For Gabor Granger, randomise the opening price across respondents so that the anchor averages out. Offer five steps, not three, and include at least one price above what you believe the ceiling is. You will be surprised more often than you expect.

That last open question is the one that repeatedly earns its place. The answers are where you find the feature or the proof point that moves the ceiling, which is a more valuable finding than the price number itself.

How do you read the output?

Plot it, then discount it, then state a range rather than a number. A single confident figure is the output most likely to be wrong.

Take a worked Van Westendorp example. Forty two responses from a mid market segment, asked per user per month. The point of marginal cheapness comes out at 28 dollars, the optimal price point at 41, the indifference price at 46, and the point of marginal expensiveness at 68. The acceptable range is therefore roughly 28 to 68, which is wide enough to be almost useless on its own.

Now overlay the Gabor Granger results from the same respondents. At 39 dollars, 71 percent said they would buy. At 49, 52 percent. At 59, 31 percent. At 69, 14 percent. Multiply acceptance by price to get relative revenue per respondent: 27.7 at 39 dollars, 25.5 at 49, 18.3 at 59, 9.7 at 69. Revenue peaks near the bottom of the range and falls off sharply above 50.

That is a much more actionable picture. The credible corridor is 28 to 68, and revenue optimises somewhere between 39 and 49. Now apply the haircut.

Apply a 20 percent haircut to the 39 to 49 window and you land around 31 to 39 dollars. That is your recommendation range, and it should reach the pricing meeting with the method, the sample size and the correction all stated.

What does the validation test look like?

A live cohort test on real traffic, for four to eight weeks, with a pre committed success metric. This is the step that separates research from opinion.

The position here is straightforward: a 40 response study followed by one live cohort test beats a 400 response study with no validation. Money changing hands is worth more evidence than any stated preference, and the cohort test is the only part of this process where that happens.

How do you present the recommendation?

With a range, a confidence statement and the method attached. Executives will push for a single number. Give them the number and refuse to drop the error bars.

A usable recommendation reads something like this. Set list price at 35 dollars per user per month for the mid market tier. Evidence: 42 responses across current customers, churned accounts and lost deals, Van Westendorp corridor 28 to 68, Gabor Granger revenue peak between 39 and 49, 20 percent stated versus revealed correction applied. Confidence: moderate. Sample is small and skewed toward one vertical. Validation: cohort test on North American new traffic for eight weeks with revenue per visitor as the metric.

That paragraph survives scrutiny. A slide that says “our research shows customers will pay 49 dollars” does not, and the first time a rep loses a deal on price somebody will go looking for the methodology.

Be explicit about what the study does not tell you. Willingness to pay research measures a point in time with your current product and current competitive set. It says nothing about price elasticity over a multi year contract, nothing about how a competitor’s price cut changes the picture, and nothing about whether your packaging is right.

What usually goes wrong

Four failures, all avoidable, all common.

Sampling only current customers, which caps your answer at what you already charge. Anchoring the survey by mentioning a current price. Mixing segments so the distribution becomes bimodal and the average describes no real buyer. And treating the number as a decision rather than an input, which is how teams end up defending a price nobody tested.

There is a fifth, subtler one. Running the research and then not changing anything, because a price increase feels risky and the study becomes a reason to feel informed rather than to act. If you are not prepared to move price, the research is entertainment. The price increase rollout playbook covers the execution side, including grandfathering policy and the communication sequence that keeps churn contained.

What to do next

Decide which question you are answering before you pick a method, because that choice determines everything else. Range, point estimate or packaging.

Then check whether you can realistically recruit the sample the method needs. If you can get 40 responses in one tight segment, run Van Westendorp and Gabor Granger together and validate with a cohort test. If you can only get 15, run interviews and be honest that the output is qualitative. The price testing and research tools roundup covers survey platforms and analysis options, the SaaS pricing frameworks compared guide covers how the number fits into a packaging decision, and the pricing change readiness checklist covers what to fix in billing and sales before anything ships. The wider SaaS pricing strategy hub holds the rest.