The Van Westendorp Price Sensitivity Meter: A Practical Guide for SaaS and AI Products

The Van Westendorp Price Sensitivity Meter is the fastest way to find out whether the price you are about to set is inside the range your market will accept. It takes four questions, it works with a few hundred respondents, and it produces a range rather than a single number, which is exactly what an early pricing decision needs.

It is also the most misused technique in pricing research. Teams run it, read the crossing point off the chart, ship that number, and wonder why revenue does not move. This guide covers the four questions, how to read the output correctly, what sample size you actually need, and where the method stops being useful.

What the Price Sensitivity Meter measures

Dutch economist Peter van Westendorp introduced the technique in 1976. It does not predict how many units you will sell at a given price. It measures perceived acceptable price: the band inside which buyers consider your product credibly priced, bounded below by "so cheap I doubt it works" and above by "more than I would ever pay".

That distinction matters. Van Westendorp tells you where your price belongs. It does not tell you which price inside that band maximises revenue. Confusing the two is the root of most bad decisions made with this data.

The four questions

Respondents see a clear description of the product, then answer four open-ended price questions in this order:

1. Too cheap: at what price would you consider the product so inexpensive that you would question its quality?
2. Cheap or good value: at what price would you consider the product a bargain, a great buy for the money?
3. Expensive: at what price does the product start to seem expensive, so that you would have to think carefully before buying?
4. Too expensive: at what price is the product so expensive you would not consider buying it at all?

Two rules govern the setup. The product description must be specific enough that everyone is pricing the same thing, including the unit of purchase, the billing period, and what is included. And the four answers must be internally consistent for each respondent, with too cheap below cheap, below expensive, below too expensive. Inconsistent responses are cleaned out before analysis.

How to read the four curves

Each question becomes a cumulative curve across the price range. Too cheap and expensive are plotted as inverse cumulative curves, cheap and too expensive as cumulative curves. Four intersections come out of it:

• Point of Marginal Cheapness (PMC): where too cheap crosses cheap. Below this, quality doubt starts to cost you buyers.
• Point of Marginal Expensiveness (PME): where expensive crosses too expensive. Above this, resistance rises sharply.
• Optimal Price Point (OPP): where too cheap crosses too expensive. This is the price at which the fewest people reject the product outright on price alone.
• Indifference Price Point (IDP): where cheap crosses expensive. Often read as the price the median respondent considers normal, or as an indication of what a market leader can command.

The band between PMC and PME is the range of acceptable prices. That range is the real output of the study. OPP is a useful reference point inside it, not an instruction.

Why the "optimal" price is usually not your price

OPP minimises price-based rejection. It is, by construction, close to the most conservative viable price. If you optimise for the fewest objections, you optimise away your margin.

For most SaaS and AI products the better decision sits in the upper half of the acceptable range, above IDP and below PME, provided the product is differentiated and the value story is credible. Companies with strong positioning routinely price at or above PME and lose only the segment they never wanted. What Van Westendorp gives you is the guardrails and the evidence for that argument, not the answer itself.

Sample size and segmentation

A single study needs enough respondents from your actual buying audience for the curves to be stable, and in practice that means a few hundred rather than a few dozen. Small samples produce curves that jump around and intersections that move materially when you add ten more responses.

Segment size matters more than total size. If you plan to price differently for self-serve and enterprise, or for two industries, each segment needs its own usable sample and its own set of curves. Pooling segments produces a blended band that fits nobody, and it hides the most valuable finding in this kind of research: that two groups value the same product at very different levels, which is the case for tiering.

Recruit buyers, not bystanders. Responses from people who would never purchase drag the whole distribution down, and that bias does not show up on the chart.

The five mistakes that make the output useless

1. A vague product description. If respondents imagine different products, the curves describe an average of things you are not selling.
2. The wrong billing unit. Per month, per year, per seat, and per run produce completely different numbers. Ask about the unit you will actually invoice.
3. Pooled segments. One band across mixed audiences hides the tiering opportunity and understates what your best segment will pay.
4. Reading OPP as the answer. It is the low-objection price, not the revenue-maximising price.
5. Stopping at Van Westendorp. The method establishes the range. It says nothing about demand or revenue at specific price points inside it.

Where Gabor-Granger takes over

Van Westendorp gives you a credible range. Gabor-Granger tells you what happens inside it. In a Gabor-Granger exercise, respondents face specific prices and state purchase intent at each one, which produces a demand curve and, multiplied out, a revenue curve across the range.

Run in sequence, the two methods answer the two different questions a pricing decision actually poses. Van Westendorp: where does our price belong so buyers take us seriously? Gabor-Granger: which price inside that band produces the most revenue, and how much do we give up by going one step lower?

Neither replaces the third input: your value metric. A perfect price attached to the wrong unit of value still breaks as usage and cost scale, which is the failure mode we see most often in AI products whose cost moves with consumption while their pricing still charges per seat.

Running one on your own product

A usable study is a week of work, not a quarter. Write a precise product description with the billing unit fixed. Field the four questions to a few hundred qualified buyers, split by the segments you intend to price differently. Clean out inconsistent responses, plot the four curves per segment, and read the acceptable band rather than the single crossing point. Then take the upper half of that band into a Gabor-Granger step to see what revenue does at each price.

That sequence is the core of how we set prices in a C.O.R.E. engagement: establish the acceptable range, test revenue inside it, and tie the final structure to a value metric that holds as the product scales. If you are still deciding which structure to research against, start with the guide to SaaS pricing models.

If you want that run on your own product and numbers, see how the C.O.R.E. engagement works.