How to Run: Van Westendorp Price Sensitivity Study

Introduction

To run a Van Westendorp price sensitivity study, ask each respondent four price questions about one described product (too cheap, a bargain, getting expensive, too expensive), drop out-of-order answers, plot four cumulative curves, and read the acceptable range under a named curve convention. The result shows where price perception breaks, not how many will buy. For demand, add the purchase-likelihood extension or run Gabor-Granger.

Last updated October 2026.

Why Two Decisions Matter More Than the Math

Most Van Westendorp studies fail before the first response arrives. The arithmetic is four columns and a handful of crossings. The damage comes from how the price is framed in the question and from which of two circulating curve conventions reads the result. Two teams can analyze the same file and report ranges that differ by a factor of three, as the worked example below shows.

Run the method when you need a defensible range for a product people already understand and you don’t yet have a shortlist of prices to test. It’s fast to field and widely used for bounding a price range on a familiar product. It measures price perception rather than demand, so it won’t tell you how many people buy at a given price.

This guide covers the design forks, the question wording, the sample you need, the calculation, and how to run it in Claude or ChatGPT with two copy-paste prompts. Pilot with a small sample, field one cell per product, remove respondents whose answers are out of order, and report a bootstrap interval on each price point.

What the Van Westendorp Price Sensitivity Meter Is

The Van Westendorp price sensitivity meter is a direct pricing method in which respondents supply prices instead of reacting to them. Peter van Westendorp, a Dutch economist, presented it at the ESOMAR Congress in Venice in 1976, in a paper titled "NSS Price Sensitivity Meter (PSM): A New Approach to Study Consumer Perception of Prices."

Every respondent answers the same four questions about one product. Each answer is a price, typically typed as a number. The analysis turns each question into a cumulative curve across the price axis. Two curves fall as price rises, because fewer people call a higher price cheap. Two curves rise, because more people call a higher price expensive.

Where those curves cross defines four named points:

Point of marginal cheapness
Optimal price point
Indifference price point
Point of marginal expensiveness

The first and last bound the acceptable price range. The optimal price point is where the fewest respondents object on either side. The indifference price point is where as many people call the price cheap as call it expensive, and it’s often read as the price the market considers normal.

Instead of asking whether someone would buy at a price, the method asks where their sense of value changes. That makes it a measure of perceived value, and it’s the source of both its speed and its main weakness.

The Extension Most Teams Should Add

In 1993, Newton, Miller and Smith presented an extension at the American Marketing Association’s Advanced Research Techniques Forum. It adds two purchase-likelihood questions, asked at the respondent’s own bargain and getting-expensive prices.

Calibrating those answers into probabilities turns the meter’s four perception curves into a trial curve and a revenue index. That step connects a price range to a demand estimate, and the design and analysis chapters below cover it.

When to Run a Van Westendorp Study

The clearest case is a new product, add-on or tier where nobody inside the company can yet name a plausible price range with confidence. Instead of guessing four candidate prices for a Gabor-Granger study, you let respondents draw the boundaries.

It typically fits four situations:

New products in familiar categories
Paid add-ons
Repricing reviews
Price sanity checks

It fits well with expert business buyers. A procurement lead or a team administrator often knows what adjacent tools cost and can answer four price questions without inventing a number.

It fits when the team needs a range to argue about instead of a single number. Pricing decisions commonly stall on whether a proposed figure is anywhere near reasonable, and a bounded range settles that faster than a point estimate.

And it fits early. The instrument is short enough to add to an existing study, which makes it a cheap first read before committing budget to conjoint analysis.

It also fits as a guardrail around a price the team has already chosen for other reasons, such as cost-plus or parity with a competitor. If the chosen price falls outside the acceptable range, the team learns that before launch instead of after.

A typical request that fits well sounds like this: "We’re launching an admin add-on and have no idea whether it should cost 5 or 50 per seat." The method generally struggles where respondents lack a reference frame. A consumer who has never bought anything like your product will still type a number, and the number will mean very little.

When Not to Run a Van Westendorp Study

Five situations, each with the instrument that’s actually right. The first two are the ones most likely to change what a reader does.

You need to know how many people will buy at a price. The four questions measure where perception shifts, not purchase. A price inside the acceptable range can still lose most of the market to a cheaper option.

Run Gabor-Granger instead, which asks yes or no at specific prices and returns a demand curve directly. Or add the Newton, Miller and Smith extension, and for a real answer, run an in-market price test.

You need to price a trade-off between features, tiers or packaging. The meter prices one described product in isolation. It can’t tell you whether buyers would pay more for a tier with single sign-on (SSO) than for one with more seats. Conjoint analysis is built for that, and it can estimate the value of each attribute against price.

You need a price relative to competitors. Sawtooth Software, in its explainer "Van Westendorp Pricing Model: Definition, How It Works, Examples, and More," names the "absence of competitive context" as a core limitation, because respondents evaluate your product alone. Don’t hope they mentally benchmark. Run a choice-based conjoint that places competitor alternatives in the choice set.

You need to decide which features belong in a tier. Price perception says nothing about which features drive it. MaxDiff ranks a feature list by relative importance, and its output can then shape the tiers you price. The MaxDiff how-to guide covers the design.

Respondents don’t yet understand what the product is. A price question about an unclear concept measures confusion. Run a product concept test first, then field the meter on the concept that wins. The meter suits a familiar product with an unknown price range. It doesn’t suit an unfamiliar product.

A sixth case is a scoping problem rather than a method problem. If the price is contractually set, regulated or negotiated deal by deal, a perception range will rarely move the decision, and the research budget is often better spent on win-loss interviews.

Routing a Pricing Question to the Right Instrument

Most pricing requests arrive phrased as one question and actually contain another. The routing table below maps the question a stakeholder typically asks to the instrument that answers it.

The last row is worth taking seriously. Every survey method in this table asks people about money they aren’t spending, and the critique chapter below covers what that does to the answer.

A useful habit is to write the stakeholder’s request down verbatim before choosing a method. A request such as "I need to understand customers' willingness to pay" can mean any of the six rows, and each row produces a different study.

Pick the row whose left-hand question matches the decision on the table. If two rows apply, run the cheaper instrument first and let its result decide whether the second is needed.

Study Design and the Three Forks

Three design decisions shape everything downstream. Make each one deliberately, because none can be fixed in analysis.

Fork One: Open Numeric Entry or a Price Ladder

The original method uses open entry: a blank box where the respondent types a price. The alternative is a price ladder, a closed list of prices the respondent picks from.

Open entry produces continuous curves and lets respondents set their own boundaries, which is the point of the method. It also produces round-number heaping, with answers piling up at 10, 20, 25 and 50, and occasional outliers from typos or protest answers.

A ladder removes the typos but imposes the researcher’s range on the answer. If the ladder tops out at 40 and a respondent’s true too-expensive point is 60, the ladder records 40 or forces a skip.

The best recent evidence on direct questions favors closed formats in general. Schmidt, Steiner, Krafft, Eckel and Dahl, in the International Journal of Research in Marketing in 2024, found that "open questions are less accurate in estimating real [willingness to pay] than closed questions." They also found that product category knowledge and cognitive abilities change which format wins, with open questions suiting customers of very high category knowledge.

That study tested direct willingness-to-pay questions generally, not the Van Westendorp meter specifically.

Default to open entry for expert buyers and familiar categories. The expertise finding supports it, and the method depends on respondents drawing their own boundaries.

Use a ladder for consumers facing a novel category, and set its range from a pilot with open entry. Instead of choosing the endpoints from internal opinion, let a small open-entry pilot tell you where the tails are. A pilot of 20 to 30 answers is a working convention for this, not a published threshold.

Fork Two: Plain Meter or the Newton, Miller and Smith Extension

The plain meter yields a range and two reference points. It doesn’t yield a demand curve, and it can’t say which price inside the range maximizes revenue.

The extension adds two questions. After the bargain price, ask how likely the respondent would be to buy at that price. After the getting-expensive price, ask the same.

Add the extension whenever the output will feed a revenue decision. It costs two questions, and it’s the only part of the method that approaches purchase.

Fork Three: One Product or Several

Four price questions about one product is the method. Four price questions about each of three tiers, asked in sequence, is a different and contaminated study, because every later answer is anchored on the earlier ones.

If you need ranges for several products or tiers, split the sample into separate monadic cells. Each respondent sees one product description and answers four questions about it. That multiplies the sample requirement, which the sample-size chapter makes concrete.

The Design Decision Tree

The three forks resolve into a short sequence:

Confirm the question is a price range for one understood product. If not, route to the instrument named in the previous chapter.
Decide whether the respondents are expert buyers in a familiar category. If yes, use open numeric entry. If no, pilot with open entry and field a ladder built from the pilot tails.
Decide whether a revenue decision depends on the result. If yes, add the two purchase-likelihood questions.
Count the products or tiers you need ranges for. Assign one monadic cell per product, and size each cell as its own study.
Decide which channel will carry the study, and check that the chosen channel supports numeric validation.

The final step is where platform constraints usually bite, so check it before building the instrument.

Running This in Sprig: Open numeric entry uses the Validated Text question type set to Numeric, which accepts numbers only. The documentation lists web SDK 2.51.0 as the minimum, with iOS, Android and React Native listed as "Coming soon," so an in-product study inside a native mobile app can’t run the open-entry branch today.

The documentation doesn’t say whether decimals or currency symbols are accepted, so test a decimal price before launch. The ladder branch works anywhere, using a single-select question.

Writing the Van Westendorp Questions

The four questions are short, and small wording changes move the curves. Hold the wording fixed across every wave you intend to compare.

The Product Description Block

Every respondent reads the same description before the first price question. It’s the most important piece of text in the study, because respondents are pricing the description rather than the product.

The block typically needs five elements:

Product name
What it does
What’s included
The billing unit
The currency

The billing unit is the element most often left out, and it causes the most damage. "Per seat per month," "per year for the whole team" and "one-time" produce different numbers from the same person. Name the unit in the description and again in every question.

Leave out any existing price, including the current list price of a related product. A number on screen anchors every answer that follows it, and the result becomes a measure of reaction to your anchor.

The Four Questions, Ready to Lift

This wording follows the version Sawtooth Software publishes, adapted to carry the billing unit. Replace the bracketed placeholders.

At what price per [UNIT] would you consider [PRODUCT] to be so inexpensive that you would feel the quality couldn’t be very good?
At what price per [UNIT] would you consider [PRODUCT] to be a bargain, a great buy for the money?
At what price per [UNIT] would you say [PRODUCT] is starting to get expensive, so that it’s not out of the question, but you would have to give some thought to buying it?
At what price per [UNIT] would you consider [PRODUCT] to be so expensive that you would not consider buying it?

Published sources order the questions differently, and no study located for this guide tests the order effect. Pick one order, hold it for every wave, and record it in the study plan.

The Two Extension Questions

If you’re running the Newton, Miller and Smith extension, add one question after the bargain answer and one after the getting-expensive answer. Pipe the respondent’s own price into the question text.

At [BARGAIN PRICE] per [UNIT], how likely would you be to buy [PRODUCT]?
At [EXPENSIVE PRICE] per [UNIT], how likely would you be to buy [PRODUCT]?

Use a five-point scale running from "definitely would not buy" to "definitely would buy." Keep the labels identical on both questions, because the analysis maps each label to a probability.

The Instrument Specification

The table below is the full instrument, in fielding order, with the input type each item needs.

| Order | Item | Input | Required | |:-----:|:-------------------------------------------:|:-----------------------:|:--------------:| | 1 | Screener on purchase role | Single select | Yes | | 2 | Product description block | Display text | Yes | | 3 | Too cheap | Numeric entry or ladder | Yes | | 4 | Bargain | Numeric entry or ladder | Yes | | 5 | Likelihood at bargain price | Five-point scale, piped | Extension only | | 6 | Getting expensive | Numeric entry or ladder | Yes | | 7 | Likelihood at expensive price | Five-point scale, piped | Extension only | | 8 | Too expensive | Numeric entry or ladder | Yes | | 9 | Open-text probe on the too-expensive answer | Open text | Recommended | | 10 | Segment attributes | Hidden or single select | Yes |

The open-text probe is optional in the method and valuable in practice. A single question asking what would make the product worth its too-expensive price commonly surfaces the missing feature or the misread billing unit that explains an odd curve. The advanced cross-tab guide covers coding those answers into themes.

Input Rules for Open Entry

Accept whole numbers or numbers with two decimals, and nothing else. Currency symbols, ranges such as "20 to 30" and words such as "free" all break the calculation.

Tell respondents the rule in the question helper text instead of rejecting their answer without explanation. A short line such as "Enter a number only, in [CURRENCY]" gives respondents the rule before they make the error rather than after.

Don’t force the four answers into ascending order in the survey, even where a platform allows it. Forcing the order hides respondents who misread a question, and those respondents are exactly the ones the quality-control chapter needs to find.

Curve Conventions and Scoring Choices

Two conventions for reading the chart circulate, and they give different answers from the same data. Name the one you use in every report.

The Two Conventions

Both conventions agree on two of the four points. The optimal price point is where the too-cheap curve crosses the too-expensive curve. The indifference price point is where the bargain curve crosses the getting-expensive curve. They disagree on the edges of the range.

Convention A inverts two curves. It builds a "not a bargain" curve as one minus the bargain curve, and a "not expensive" curve as one minus the getting-expensive curve.

The point of marginal cheapness is where too cheap crosses not a bargain, and the point of marginal expensiveness is where not expensive crosses too expensive. The R package pricesensitivitymeter uses this convention in its psm_analysis function. It also offers an option for a narrower acceptable range, which some research firms use.

Convention B uses the uninverted curves. The point of marginal cheapness is where too cheap crosses getting expensive, and the point of marginal expensiveness is where bargain crosses too expensive. Conjointly’s manual-calculation guide uses this convention.

What the Difference Looks Like

The worked example used throughout this guide is a hypothetical analytics add-on priced per seat per month, fielded to 320 respondents, with 300 kept after cleaning. All figures are illustrative and describe no real product.

| Point | Convention A | Convention B | |:-------------------------------:|:------------:|:------------:| | Point of marginal cheapness | 11.25 | 15.00 | | Optimal price point | 16.00 | 16.00 | | Indifference price point | 19.25 | 19.25 | | Point of marginal expensiveness | 27.00 | 20.25 | | Width of the acceptable range | 15.75 | 5.25 |

The same 300 people produce an acceptable range three times wider under one convention than the other. A team reading Convention B would conclude that anything above about 20 is risky, while a team reading Convention A would see headroom to 27.

Neither convention is an error. Convention B produced the narrower range here, which some teams prefer as the conservative read. Convention A produced the wider range, bounded by where rejection starts to dominate.

Report Convention A as the primary range and Convention B alongside it, labeled. Convention A is primary because it follows the original definition and the package default, but it’s the more permissive read, so treat any price above Convention B’s upper edge as contested. Don’t choose the convention after seeing which answer the room prefers. Fix it in the study plan before fielding.

The Purchase-Likelihood Scale

For the extension, each point on the five-point scale needs a probability. The pricesensitivitymeter package defaults to 70 percent for "definitely would buy," then 50, 30, 10 and 0 percent down the scale.

Those values are a software default rather than a published finding. Treat them as a starting assumption, and if your category has historical survey-to-purchase conversion data, calibrate from that instead.

Sample Size and Precision

No published source derives a Van Westendorp sample-size rule. Vendor pages commonly quote a respondent count, and none located for this guide shows the arithmetic behind it.

So derive it. Each of the four curves is a cumulative proportion: the share of respondents whose answer falls above or below a given price. The precision of a crossing point is therefore governed by the margin of error on a proportion near that point. The NPS guide walks through the same arithmetic for a score.

The Derived Precision Table

At 95 percent confidence, the margin of error on a proportion is 1.96 times the square root of p times one minus p, divided by n, all under the square root. The worst case is a proportion of 0.5.

Crossings at low heights, such as the optimal price point in the worked example near 0.03, are where this table says least and the bootstrap below matters most.

This table is derived, not sourced. It shows how uncertain each curve height is, which is the input to how uncertain each crossing is.

Why the Table Understates the Uncertainty

A crossing point moves when either of two curves moves, and flat curves move the crossing a long way for a small change in height. Where two curves meet at a shallow angle, a few points of noise can shift the crossing by several price units.

That’s why the defensible way to report precision is a bootstrap interval on each price point, not a margin on a proportion.

The Bootstrap Recipe

Take the cleaned respondent file.
Draw a sample of the same size with replacement.
Recompute all four price points using your chosen convention.
Repeat at least 1,000 times.
Report the 2.5th and 97.5th percentiles of each point as its 95 percent interval.

On the worked example, 1,000 resamples produced these intervals under Convention A:

| Point | Estimate | 95 percent interval | |:-------------------------------:|:--------:|:-------------------:| | Point of marginal cheapness | 11.25 | 10.25 to 12.25 | | Optimal price point | 16.00 | 15.25 to 17.25 | | Indifference price point | 19.25 | 18.00 to 20.00 | | Point of marginal expensiveness | 27.00 | 25.00 to 28.00 |

The upper edge of the range carries the widest interval in this example, because the too-expensive answers spread out more than the too-cheap ones.

Segments and Cells

The requirement applies to every cell you report. Three tiers fielded as monadic cells need three full samples, and a segment cut by company size needs its own sample within each cell.

Don’t choose a sample total and divide it. Decide the widest interval the decision can tolerate, then size each cell to reach it. If a pricing committee can act on a range edge known to within about two price units, the margins in the table above say whether 200 per cell gets there or whether you need 400.

A bootstrap on a 20-to-30-person pilot is too unstable to extrapolate, so use the pilot only to check the question wording and the price range.

Audience and Screening

Screen on purchase role, not on product usage. The person who uses a tool daily and the person who approves its invoice frequently hold different price boundaries, and the approver’s are usually the ones that matter.

For business products, a single screener typically does the work: "Which best describes your role in purchasing software like this?" with options for final decision, significant influence, some input and none. Keep the first two, and field the third as a separate segment if its size allows.

For consumer products, screen on category purchase in a stated window, such as having bought a comparable product in the past 12 months. Someone who has never paid for the category has no boundary to report.

Customers or the Market

Existing customers and non-customers answer the meter differently, and both answers are useful for different decisions.

Existing customers anchor on what they currently pay. Their curves are informative for an add-on or a renewal price and misleading for acquisition pricing, since they typically cluster near the incumbent price.

Non-customers from a panel describe the market you’re trying to enter. Their curves are the right input for a launch price, provided the screener holds them to real category buyers.

Don’t blend the two groups into one curve. Field them as separate cells or tag them as a segment. A blended curve describes nobody.

Attributes to Include

Attach segment attributes before launch, because attributes not recorded at response time generally can’t be reconstructed from an export. For business products, company size, current plan and seat count are commonly the cuts a pricing committee asks for first.

Running This in Sprig: In-product targeting reaches existing customers at a relevant moment, such as a visit to the billing page, and panels reach non-customers against screening criteria.

Response quotas are documented as Enterprise-only, and their screener questions are limited to Single Select, Rating Scale and NPS, so a quota can hold a segment cell but can’t hold a price band. Panel respondents aren’t your buyers until the screener proves it, so budget for a higher screen-out rate on business audiences.

Fielding the Study

The meter is short, six questions with the extension, which makes it flexible about channel. The channel still changes who answers.

Email and link surveys reach a defined list and allow a full product description, which suits launch pricing and repricing studies. In-product surveys reach active users in context, which suits add-on pricing, as long as the description block still carries the full offer.

Keep the meter in its own study or at the end of a longer one, never in the middle. Questions about value earlier in a survey, such as satisfaction or feature ratings, can shift the price answers that follow, and ordering it last contains that effect.

Pilot with 20 to 30 respondents before the full launch. Read the raw answers, not the curves. The pilot is where you discover that people are answering per year when you meant per month, and that discovery is free at 30 respondents and expensive at 300.

For a ladder design, the pilot is also where the ladder endpoints come from. Pad the ends, because a small pilot understates the tails. Set the bottom rung at about half the lowest pilot too-cheap answer and the top rung at about twice the highest pilot too-expensive answer, and add an "Above this range" option.

Space the rungs evenly on a ratio scale rather than an arithmetic one. Rungs at 5, 7, 10, 14, 20, 28 and 40 give equal resolution at every price level, while rungs every 5 waste resolution at the top and lack it at the bottom.

International and Multi-Currency Studies

Field one cell per market and currency, with the description block localized and not translated word for word. Price perception is commonly tied to local reference prices, so a converted range from one market is rarely a usable estimate for another.

Where the business needs a single global range, report each market separately first, then state the conversion rate and date used for any combined view.

Incentives

Keep any incentive unrelated to the product. Avoid offering a discount on the product being priced, since it tells respondents what the company thinks the product is worth.

The Pre-Launch Checklist

Work through these in order before the first invitation goes out.

Write the pricing decision the study serves, in one sentence, and name who makes it.
Choose the curve convention and the question order, and record both in the study plan.
Draft the product description block, with the billing unit and currency named.
Remove every existing price from the description, the screener and the invitation.
Confirm the channel supports numeric validation, and test a decimal price on a real device.
Set the screener to purchase role, and decide how to treat each answer option.
Fix the cell count, and size each cell from the widest interval the decision can tolerate.
Pipe the respondent’s own prices into the two extension questions and preview the piping.
Write the exclusion rules, including whether ties count as consistent, before any data exists.
Add the segment attributes you’ll need for the cuts a committee will ask for.
Run a pilot of 20 to 30 and read the raw answers for a misread unit.
Check that the export includes every answer column with a short, plain name.
Set the speed floor from the pilot median.
Schedule the readout against the date the price is decided, not the date fielding ends.

Timing and Cadence

Run the meter when a pricing decision is actually open. Price perception typically moves slowly for an established product, and refielding an unchanged offer every quarter mostly returns the same range with noise on top.

What justifies a refield is a change in the offer or the market: new packaging, a competitor repricing, a material feature release, or a currency shift for international pricing. Hold the description block, the wording and the question order identical, or the waves stop being comparable.

Avoid fielding immediately after a price announcement. Respondents who just saw a new price typically report boundaries anchored on it.

Allow enough calendar time for the decision rather than for the fieldwork. The pilot, the cleaning rules and the committee readout frequently take longer than the fielding itself.

Don’t schedule the meter as a recurring tracker. Attach it to the pricing calendar. A study that finishes a week after the price is set is a retrospective.

Quality Control and Data Hygiene

Van Westendorp data fails in four predictable ways, and each has a mechanical check.

Out-of-order answers. A respondent whose bargain price exceeds their getting-expensive price has misread a question or answered carelessly. Test every row for too cheap at or below bargain, bargain at or below getting expensive, and getting expensive at or below too expensive.

Decide in the study plan whether ties count as consistent. The worked example kept ties, which retained 300 of 320 respondents. A strict rule would have kept 287. Report the count under the rule you chose.

Implausible values. Zeros, single-digit answers for an enterprise product and seven-digit answers are typically typos or protests. Flag anything below the pilot minimum or above the pilot maximum by a wide margin, review those rows by hand, and remove them only with a written rule.

Speeders and bots. A respondent who completes four price questions in a few seconds hasn’t read the description block. Set a speed floor from the pilot median, and review rather than delete automatically.

Heaping. Count the share of answers that land on multiples of 5 or 10. A share far above what the price scale would produce by chance is normal for open entry, and it matters mainly for interpretation, because it makes crossings snap to round numbers.

Out-of-order answers are an automatic exclusion under a written rule. For implausible values and speed, review rows that fail two or more checks, because a single out-of-range answer is often a typo in an otherwise careful response.

Report the funnel every time, before anyone sees a range: respondents fielded, removed for ordering, removed for implausible values, removed for speed, and kept.

Running This in Sprig: Bot detection is documented for survey links and is off by default. It labels likely bots "Suspected bot" and hides them from displayed results through a toggle. It doesn’t delete them, and CSV exports still include them, so filter them out before analysis.

The documentation doesn’t say whether in-product surveys are covered, and the feature doesn’t check the four answers against each other. The ordering test above runs on the export or in the AI client.

The Core Calculation

The calculation turns four columns of prices into four cumulative curves, then finds where specific pairs of curves cross. It’s simple arithmetic, and it’s easy to get subtly wrong.

Building the Four Curves

Sort every unique price that appears in any of the four columns. At each price, compute four shares across the cleaned respondents:

The first two curves fall from left to right. The last two rise. For Convention A, add the two inverted curves: not a bargain equals one minus bargain, and not expensive equals one minus getting expensive.

Finding the Crossings

A crossing is the price at which one curve moves from above another to at or below it. On real data the curves are step functions, so two choices matter.

The first is the price grid. Evaluating only at prices respondents actually typed produces crossings that jump between round numbers. Evaluating on a fine grid, such as every 0.25 or every 0.01, and interpolating between steps gives a stable point.

The second is how to handle a tie across a flat stretch. When two curves overlap for a range of prices, report the first price where the crossing occurs and say so. The pricesensitivitymeter package exposes an intersection-method option for exactly this reason.

The Worked Example at Four Prices

Here are the curve heights from the worked example at four prices, under the 300 kept respondents.

| Price per seat per month | Too cheap | Bargain | Getting expensive | Too expensive | |:------------------------:|:---------:|:-------:|:-----------------:|:-------------:| | 12 | 0.180 | 0.763 | 0.037 | 0.007 | | 16 | 0.030 | 0.407 | 0.203 | 0.037 | | 20 | 0.003 | 0.263 | 0.403 | 0.150 | | 27 | 0.000 | 0.040 | 0.733 | 0.303 |

Read the 16 row against the 20 row. The bargain curve falls from 0.407 to 0.263 while the getting-expensive curve rises from 0.203 to 0.403, so they cross between the two prices.

On the full 0.25 grid, which carries many more rows than this excerpt, the crossing lands at 19.25, which is the indifference price point. Interpolating linearly between only these four rows would give about 18.4, which is why the grid matters.

Adding the Purchase-Likelihood Extension

For each respondent, the extension produces four anchor points on a personal purchase-probability curve:

Zero probability at or below the too-cheap price
The calibrated probability at the bargain price
The calibrated probability at the getting-expensive price
Zero probability at or above the too-expensive price

Interpolate linearly between those anchors, then average across respondents at each price to get a trial curve. Multiply the trial curve by price to get a revenue index.

The price that maximizes trial and the price that maximizes the revenue index are typically different, and the gap between them is often the most useful number the study produces. Treat both as relative indices rather than forecasts. The calibration is an assumption, and the whole curve inherits it.

Benchmarks and Why There Are None

There’s no benchmark for a Van Westendorp result, and there couldn’t be one: the four price points are properties of the product description you showed, the audience you asked and the currency they answered in, so another company’s optimal price point tells you about their study, not your market.

That holds within a company too. Two studies with different description blocks, billing units or screeners produce ranges that can’t be compared, even for the same product.

Vendor pages frequently publish rules of thumb about how wide an acceptable range should be, or where the optimal point should sit relative to the indifference point. None located for this guide cites a source, and none should be used to grade a result.

What You Can Compare

Three comparisons are legitimate, all within your own program.

Wave against wave, with an identical description, wording, question order, screener and convention. A range that shifts upward after a feature release is a real signal when everything else held still.

Cell against cell within one study. Two tiers fielded as monadic cells to the same screened population can be compared directly, and the difference between their ranges is a measure of how much more the second tier is perceived to be worth.

Segment against segment within one cell, provided each segment has enough respondents for its own bootstrap interval. A mid-market segment whose too-expensive curve sits well to the right of the small-business curve is a pricing argument for separate plans.

What Not to Compare

Don’t compare your range with a range quoted in a vendor case study, an analyst report or a competitor’s published research. Those studies typically used a different description, a different audience and an unstated convention, so the comparison measures three study-design differences at once.

Don’t compare the meter’s range with a Gabor-Granger or conjoint result for the same product either. Those methods answer different questions, and agreement between them is reassuring while disagreement is expected rather than alarming.

Build Your Own Reference Series

Instead of looking for an external norm, keep a log of every study: the date, the description block verbatim, the audience, the convention, the four points and their intervals. After three or four studies on the same product, you have the only benchmark that will ever apply to it.

Interpreting and Acting on the Result

Start with the range, not the points. The acceptable range answers the question a pricing committee usually asks first: whether a proposed price is anywhere near reasonable.

A proposed price inside the range under both conventions is defensible on perception grounds. A price inside Convention A’s range but outside Convention B’s sits in contested territory, and the report should say so plainly.

What Each Point Licenses You to Say

The optimal price point is the most misread number the method produces. It minimizes objection, which isn’t the same as maximizing trial or revenue. Where revenue is the goal, the purchase-likelihood extension, not the optimal point, is the part of the study that addresses it.

Reading the Width

A narrow range means respondents agree about what the product is worth. A wide range means they don’t, which is often the more interesting finding.

In the worked example, the range under Convention A runs from 11.25 to 27.00. That width can point to two audiences hiding in one sample, and the segment cut is where to look next.

Turning the Result Into a Decision

Write the recommendation as a conditional pair rather than a single number. If the goal is the least objection, price near the optimal point, 16 in the worked example.

If the goal is adoption or revenue and the extension ran, price near the trial peak or the revenue-index peak, provided either sits inside the acceptable range.

Present both conventions on the same chart, with the proposed price drawn as a vertical line. A committee can then see at a glance whether the proposal sits comfortably inside both ranges, inside one, or outside both.

Then state what the study can’t tell the committee, which is how buyers will respond when the price appears next to a competitor’s. Route that question to conjoint analysis or a live test.

Running the Analysis With Claude or ChatGPT

A general-purpose AI client with code execution can run the whole Van Westendorp calculation from a CSV export. Researchers remain responsible for the numbers that leave the conversation.

The risk isn’t that the model can’t do the arithmetic. The risk is that a vague prompt lets it choose a convention silently, skip the ordering check, or estimate a crossing by eye.

What the Analysis Produces

Two outputs, one per prompt. The first is the four price points under both conventions, with bootstrap intervals, a cleaning funnel and a chart. The second is the trial and revenue-index curves from the purchase-likelihood extension.

Both prompts share eight conventions: a stated purpose and output, code instead of estimation, an explicit exclusion rule, a threshold that names zero, an independent recompute by a different method, an instruction to report mismatches, a named output file, and bracketed placeholders with defaults.

Prompt One: The Four Price Points

What this prompt does: cleans the four price columns, builds the cumulative curves, and finds the four price points under both conventions with bootstrap intervals.

What it returns: a cleaning funnel, a results table and a chart saved as a named file.

What this prompt does: computes Van Westendorp price points under two
conventions with bootstrap intervals.
What it returns: a funnel, a results table and a chart.

Use code to calculate this, not estimation.

Data: the attached file [FILE NAME, default: van_westendorp_export.csv].
Columns: [TOO CHEAP COLUMN, default: too_cheap], [BARGAIN COLUMN,
default: bargain], [EXPENSIVE COLUMN, default: expensive], [TOO
EXPENSIVE COLUMN, default: too_expensive], and optionally [SEGMENT
COLUMN, default: none].
Unit: [PRICE UNIT, default: dollars per seat per month].

Step 1. Exclusions. Drop any row where any of the four price columns is
blank, non-numeric, zero, or negative, or where any answer is below
[MIN PRICE, default: pilot minimum divided by 2] or above [MAX PRICE,
default: pilot maximum times 3]. Then drop any row where the
answers are out of order, using this rule: too cheap <= bargain <=
expensive <= too expensive [TIES RULE, default: ties allowed].
Report a funnel: rows received, dropped for blanks, dropped for
non-numeric or zero, dropped for order, dropped for implausible values, rows kept.

Step 2. Threshold. If rows kept in the full sample or in any segment
is below [MINIMUM ROWS, default: the minimum base in your study plan],
say so and don’t report that group’s price points. A segment with
zero rows counts as below the threshold and must be listed.

Step 3. Curves. On a grid from the lowest to the highest price in the
data at a step of [GRID STEP, default: 0.25], compute: share with too
cheap >= price, share with bargain >= price, share with expensive <=
price, share with too expensive <= price. Also compute not bargain =
1 - bargain and not expensive = 1 - expensive.

Step 4. Price points. A crossing is the first grid price where the
first curve moves from above the second to at or below it.
Convention A: marginal cheapness = too cheap x not bargain. Optimal =
too cheap x too expensive. Indifference = bargain x expensive.
Marginal expensiveness = not expensive x too expensive.
Convention B: marginal cheapness = too cheap x expensive. Marginal
expensiveness = bargain x too expensive. Optimal and indifference as
in A.

Step 5. Intervals. Bootstrap the kept rows [RESAMPLES, default: 1000]
times with replacement, recompute every point under both conventions,
and report the 2.5th and 97.5th percentiles.

Step 6. Recompute. Independently recompute the four Convention A
points by a different method: sort each column and derive each curve
from rank positions rather than from the grid comparison in Step 3.
Report both results side by side.

Step 7. Honesty. If Step 6 disagrees with Step 4 by more than one grid
step on any point, report the mismatch and both values. Don’t correct
either silently.

Output: a table of points, estimates and intervals for both
conventions, the funnel from Step 1, and a chart of the four curves
with the points marked, saved as [CHART FILE, default:
psm_curves.png].

The minimum base in Step 2 is a placeholder, not a published floor. Set it from the interval your decision can tolerate, using the derived precision table above.

Prompt Two: The Purchase-Likelihood Extension

What this prompt does: converts the two likelihood answers into per-respondent purchase curves, then aggregates them into trial and revenue-index curves.

What it returns: a table of trial and revenue index by price, the two peak prices, and a chart saved as a named file.

What this prompt does: builds trial and revenue-index curves from the
two purchase-likelihood answers.
What it returns: a table, two peak prices and a chart.

Use code to calculate this, not estimation.

Use the cleaned file from the first prompt, and reattach it if this is a new conversation. Additional columns:
[LIKELIHOOD AT BARGAIN COLUMN, default: likelihood_bargain] and
[LIKELIHOOD AT EXPENSIVE COLUMN, default: likelihood_expensive],
scored 1 to 5 where 5 = definitely would buy.

Exclusions: drop any kept row where either likelihood answer is blank
or outside 1 to 5. Report how many rows this removes. If fewer than
[MINIMUM ROWS, default: the minimum base in your study plan] remain,
or zero remain, stop and say so.

Calibration: map 5, 4, 3, 2, 1 to [CALIBRATION, default: 0.70, 0.50,
0.30, 0.10, 0.00]. State in the output that these values are an
assumption.

For each respondent, set purchase probability to 0 at or below too
cheap, the calibrated value at bargain, the calibrated value at
expensive, and 0 at or above too expensive. Interpolate linearly
between those four points.

Average across respondents at each price on a grid of [GRID STEP,
default: 0.25] to get trial. Multiply trial by price to get a revenue
index. Report the price that maximizes each.

Recompute: calculate trial at the two peak prices a second way, by
looping over respondents one at a time with no vectorized code, and
report both values. If they differ by more than 0.005, report the
mismatch rather than choosing one.

Output: a table of price, trial and revenue index, the two peak prices,
and a chart saved as [CHART FILE, default: psm_trial_revenue.png].

Setup and Connecting Your Data

Enable code execution in the client before starting. Both Claude and ChatGPT will otherwise attempt the arithmetic in text, which is exactly the failure the first line of each prompt forbids.

Export one row per respondent with the four price columns, the two likelihood columns if the extension ran, and any segment attributes. Rename the columns to short plain names before upload, and keep the original file untouched. If the export includes a bot score, filter on it first.

For more than a few thousand rows, keep the file as an attachment instead of pasting it into the message. Pasted data above the instructions tends to push the instructions out of focus.

Running This in Sprig: The Sprig Model Context Protocol (MCP) server lets a connected Claude or ChatGPT read study responses directly, up to 1,000 per call according to the documentation, which skips the export step.

A connected client can draft a study but can’t launch one. The calculation still runs in the AI client under the eight prompt conventions, so check the funnel and the recompute before the numbers reach a deck. For a study over 1,000 responses, export instead.

Reading the Output

Check the funnel first. If noticeably more rows fall out on ordering than fell out in the pilot, reread the question wording for a misleading phrase before trusting any curve.

Then compare the two conventions. A large gap between them, as in the worked example, isn’t a model error. It’s a property of how spread out the bargain and getting-expensive answers are, and it belongs in the report.

Then read the intervals. Any point whose interval spans more than the price step the committee is choosing between can’t settle that choice.

What to Verify Before Reporting

Confirm the recompute in Step 6 matched. Confirm the chart marks the same prices as the table. Confirm the segment list includes every segment, including any with zero rows. Confirm the calibration values printed match the ones you intended.

Then run the first prompt a second time in a fresh conversation. The consumer chat interfaces don’t offer a seed setting, and research from Thinking Machines Lab in September 2025 found that 1,000 completions at temperature zero produced 80 unique outputs.

That test ran an open-weight model on an inference server, not the consumer clients, but the lesson transfers. A reportable number should be produced twice.

Pitfalls

Silent convention choice. A prompt that says "calculate the Van Westendorp price points" gets whichever convention the model saw most often. Name both, every time.

Grid artifacts. A coarse grid makes crossings snap to round numbers. If every point lands on a multiple of five, the grid is too coarse or the data is heaped, and the report should say which.

Confirmation prompts. Asking the model whether 20 looks like a good price invites agreement. Sharma and colleagues, at ICLR in 2024, documented this sycophancy across assistants. Ask for the calculation, never for validation of a number you already favor.

Lost instructions. Liu and colleagues, in Transactions of the Association for Computational Linguistics in 2024, found that models use information at the start and end of a long context better than information in the middle. Keep the prompt short and the data in a file.

Open-text probes. If you theme the open-text answers on the too-expensive question, remember that applying an existing codebook is an easier job for a model than discovering themes. Hill and colleagues, in PLOS Digital Health in April 2026, tested one healthcare focus-group transcript.

Models matched human coders when applying a codebook and were weaker on latent themes. One transcript is a narrow base, so treat the result as a caution rather than a rule.

Theme once yourself, then have the model apply your list, as the advanced cross-tab guide describes.

Personal information in verbatims. Open-text answers sometimes contain names, company details or contract terms. Check which tier of the AI client you’re on, because the data-use terms depend on the tier rather than the vendor. The guide on analyzing survey data with AI covers disclosure and validation.

The Critique You Should Know

Three objections recur in the literature, and one study cuts both ways. A reader presenting a Van Westendorp range to a pricing committee should expect all three.

Hypothetical Answers Overstate Willingness to Pay

The general objection to direct pricing questions is that respondents aren’t spending anything. Miller, Hofstetter, Krohmer and Zhang tested this directly in the Journal of Marketing Research in 2011, volume 48 issue 1, pages 172 to 184, in a paper titled "How Should Consumers' Willingness to Pay Be Measured? An Empirical Comparison of State-of-the-Art Approaches."

They compared four approaches against real purchase data: an open-ended question, choice-based conjoint, the Becker, DeGroot and Marschak incentive-compatible mechanism, and incentive-aligned conjoint. Respondents were more price sensitive when real money was at stake, and the incentive-compatible mechanism and incentive-aligned conjoint passed both the statistical and the decision-oriented tests.

The authors also suggest that the open-ended and hypothetical conjoint formats may show hypothetical bias yet still yield accurate demand curves and pricing decisions. That’s a counterpoint inside the same paper: a biased level isn’t always a wrong decision.

State the scope limit plainly. That study tested a single open-ended willingness-to-pay question, not the four-question meter. The concern transfers by analogy, and its magnitude doesn’t.

The Crossings Have No Theory Behind Them

The second objection is structural. Sawtooth Software, whose tools are commonly used to run the method, writes that the line-crossing chart "doesn’t follow from economic theory."

Nothing in consumer theory says the price where too cheap meets too expensive should be optimal in any economic sense. The label is a naming convention from 1976, and it typically carries more authority in a deck than the arithmetic behind it supports.

This is practitioner critique from a vendor rather than peer-reviewed work, and it’s worth labeling as such when you cite it.

A related point concerns what the meter optimizes. Kloss and Kunter, discussed below, describe the method as biased by its hypothetical nature. A price chosen to minimize objection is a defensible goal, but it’s a different goal from maximizing revenue or margin, and a committee that hears "optimal" typically assumes the latter.

Open Questions Are Less Accurate Than Closed Ones

The third objection comes from the format fork in the design chapter. Schmidt and colleagues, in the International Journal of Research in Marketing in 2024, volume 41 issue 2, pages 383 to 402, found open questions less accurate than closed questions at estimating real willingness to pay.

Again, that study tested direct questions generally. It didn’t field the Van Westendorp meter.

The Study That Cuts Both Ways

Kloss and Kunter compared the meter itself against contingent valuation and the Becker, DeGroot and Marschak mechanism, in the European Journal of Management in 2016, volume 16 issue 2, pages 45 to 54, in a paper titled "The Van Westendorp Price-Sensitivity Meter as a Direct Measure of Willingness-to-Pay."

Their finding holds both sides of the argument at once. The meter is biased because of its hypothetical nature, they report, yet its results were comparable to those of the incentive-aligned Becker-DeGroot-Marschak mechanism and showed high predictive quality.

Weigh that correctly. It’s one study, and it shouldn’t be quoted as settling the question. What it does establish is narrower: in at least one direct test, the meter’s bias didn’t stop it from tracking an incentive-aligned benchmark.

The Counter-Critique From Expertise

Schmidt and colleagues also found that open questions suit customers with very high category knowledge. That’s a moderator, not a rebuttal, and it lands on exactly the audience where the meter is most commonly used: business buyers who already know what comparable tools cost.

The Position This Guide Takes

The Van Westendorp meter is a widely used instrument for bounding a price range, though the validity evidence is thin and mixed, and an unreliable one for predicting demand inside it. Use it to find where perception breaks, use the extension or Gabor-Granger to estimate response inside that range, and use a live test when the revenue at stake justifies one.

In practice, that means presenting the acceptable range with its intervals, labeling the optimal price point as the point of least objection, and naming the follow-up instrument on the same slide. Don’t defend the method against its critics. Scope it to the question it can answer.

Common Mistakes

Leaving the billing unit out of the questions. Respondents answer per month, per year or per seat according to their own assumption. The curves blend incompatible answers, and nothing in the output flags it. Name the unit in the description and in all four questions.

Showing a current price. Any number on screen anchors every answer. The study then measures reaction to your anchor rather than perceived value, which is often the opposite of what the team wanted to learn.

Asking four questions about several tiers in sequence. Every later tier is priced relative to the earlier one. Split tiers into monadic cells, even though it multiplies the sample.

Forcing ascending order in the survey. Validation that blocks out-of-order answers hides the respondents who misread a question. Let them answer, then remove them by a written rule in analysis.

Not naming the curve convention. Two analysts reading the same file can report ranges that differ by a factor of three, as the worked example shows. Put the convention in the study plan and on every chart.

Reading the optimal price point as the revenue-maximizing price. The point minimizes objection. It frequently sits below the price that would earn more, and the extension exists to find that price.

Reporting points without intervals. A point of marginal expensiveness reported as 27.00 invites a debate between 26 and 28 that the data can’t settle. Report the bootstrap interval every time.

Blending customers and non-customers. Existing customers anchor on what they pay today. A blended curve describes neither the base nor the market.

Comparing across studies that changed the description. A new description block is a new study. Waves are only comparable when everything upstream of the answers held still.

Placing the price questions before the description. A respondent who meets the first price question before reading what the product includes will guess, and the guess becomes a data point. Show the full description block first, on its own screen if the channel allows.

Mixing currencies in one curve. An international study that pools answers in euros, pounds and dollars produces a curve in no currency at all. Field one cell per currency, or convert with a stated rate before building any curve, and say which you did.

Reading heaped answers as signal. Round numbers attract answers, so the curves often jump at 10, 20 and 50. A crossing that lands exactly on a round number usually reflects that heaping, not a real threshold, and the interval will generally show it.

Dropping inconsistent respondents without reporting it. Removing out-of-order rows is correct. Removing them silently isn’t, because a high removal rate is itself a finding about the question wording.

Skipping the pilot. A small pilot typically reveals a misread billing unit or a confusing question in an afternoon. Discovering the same problem after 300 responses means refielding the whole study.

Treating a wide range as a failed study. A wide range typically means two audiences with different boundaries. Cut by segment before concluding anything, which is what cross-tab analysis is for.

Synthetic Respondents

Don’t field the meter to simulated respondents and report the range as market perception.

The strongest evidence on simulation is Brand, Israeli and Ngwe, a Harvard Business School working paper first released in March 2023 and last revised in April 2026. They report that willingness-to-pay estimates from language model responses are sometimes comparable to human studies, but often inaccurate, and in some cases the sign is wrong.

The same paper reports the boundary that matters here. Fine-tuning on prior survey data improved alignment for new features within the same category and population, with no similar improvement for new categories or for differences between customer segments. Those are cases a pricing team can easily face. The authors conclude that language models should supplement human studies, not replace them.

There’s also a structural reason for caution. A language model asked for a too-expensive price may draw on published prices it has seen, which makes its answer a summary of list prices rather than a measure of anyone’s boundary. For a new category, there may be no list prices to draw on at all.

Where simulation generally helps is on the instrument, with a researcher reviewing every flag. Ask a model to read the description block and the four questions and restate the billing unit it assumed. Any mismatch with your intent is a mismatch your respondents will commonly share.

How Sprig Supports Van Westendorp Studies

Sprig has a Van Westendorp pricing template rather than a built-in question type, and teams adapt the description block and wording to their product. Open numeric entry uses the Validated Text question type, which is web-only today, and a single-select ladder works on every channel.

Distribution typically follows the audience:

Web in-product surveys for existing customers
Email to customer lists
Shareable links
Panels for non-customers

The Analyze Agent lists Van Westendorp among its analyses, alongside crosstabs, t-tests, ANOVA and correlation, to explore price sensitivity and identify an acceptable price range.

The prompts in this guide work in any connected AI client, and they name the curve convention and the bootstrap intervals explicitly. The Analyze Agent page describes what it covers. The Synthesize Agent themes the open-text probe, and researchers remain responsible for reviewing those themes against the verbatims before any of them reaches a report.

Where Other Tools Lead

Two examples show that survey platforms ship Van Westendorp as built-in analysis, and a buyer comparing tools should know it. As of October 15, 2026, SurveyMonkey’s LaunchPad help page describes a Van Westendorp solution in its LaunchPad product with four standard, non-editable questions, a price sensitivity meter output with the four points and the acceptable range, filtering by custom questions, and export to XLSX, SPSS or CSV.

The page lists the solution as an optional add-on, says it requires data stored in the US data center, and doesn’t state which plans include it.

As of the same date, QuestionPro’s help page describes a Van Westendorp question type with editable question text, minimum and maximum prices, reversible question order and a dashboard chart. The page lists the feature under the Research Edition license.

For a team that wants the chart and the acceptable range produced inside the survey tool with no analysis step, those tools are the shorter path. A team that wants an explicit convention, bootstrap intervals and the extension’s revenue index can run the prompts above on an export from any of them.

Alternatives and Adjacent Methods

Each alternative answers a question the meter can’t, and several pair well with it.

Gabor-Granger asks yes or no at a set of prices you choose and returns a demand curve. Use it once the meter has told you where the plausible range sits, and pick your test prices from inside that range.

Conjoint analysis estimates how price trades against features and can include competitor alternatives in the choice set. It’s often the second study after a meter, once the question moves from "what range" to "which package."

MaxDiff ranks features by relative importance. It feeds packaging decisions that the meter then prices.

An in-market price test measures real purchase. It’s the only method here without hypothetical bias, and it’s typically the slowest and the riskiest to run.

Win-loss interviews explain why deals were lost on price, which is often a different reason from the one the price question implies. Instead of inferring the reason from a curve, ask the buyers who walked away.

A product concept test comes before any of these when the offer itself is still unsettled. Pricing a concept that respondents misunderstand produces a precise range for the wrong product.

The pairing most teams end up with is sequential rather than either-or:

Run a concept test to confirm the offer is understood.
Run the Van Westendorp meter to find the acceptable range.
Run Gabor-Granger or conjoint on candidate prices inside that range.
Run a live test on the finalist price where the revenue at stake justifies it.

Not every decision needs all four steps. A small add-on often stops after step two, while a new flagship tier commonly justifies step four.

Frequently Asked Questions

What Are the Four Van Westendorp Questions?

The four Van Westendorp questions ask at what price a product becomes too cheap to trust, a bargain, getting expensive, and too expensive to consider. Each answer is a price the respondent types or selects. The full wording, ready to lift, is in the chapter on writing the questions above, and the conjoint how-to guide covers the next study when the question moves to packaging.

How Many Respondents Does a Van Westendorp Study Need?

Aim for roughly 200 to 400 respondents per cell, because no published source derives a Van Westendorp sample-size rule and the precision follows from the arithmetic.

Each curve is a cumulative proportion, so 100 respondents per cell puts each curve height within about 10 points at 95 percent confidence, 200 within about 7, and 400 within about 5. Report a bootstrap interval on each price point rather than relying on the count alone.

What Are the Four Van Westendorp Price Points?

The four Van Westendorp price points are the point of marginal cheapness, the optimal price point, the indifference price point and the point of marginal expensiveness. The first and last bound the acceptable price range. The optimal price point is where the fewest respondents object on either side, and the indifference price point is where equal shares call the price a bargain and getting expensive.

How Do You Calculate the Van Westendorp Price Points?

The Van Westendorp price points come from four cumulative curves. Compute, at each price, the share calling it too cheap or a bargain at or above that price, and the share calling it getting expensive or too expensive at or below it. The optimal price point is where too cheap crosses too expensive, and the indifference price point is where bargain crosses getting expensive.

What Is the Difference Between the Optimal Price Point and the Indifference Price Point?

The optimal price point is where the fewest respondents object that the price is too cheap or too expensive. The indifference price point is where equal shares call the price a bargain and getting expensive, and it’s often read as the normal market price. Neither is the revenue-maximizing price.

What Is the Newton, Miller and Smith Extension?

The Newton, Miller and Smith extension adds two purchase-likelihood questions to the meter, asked at the respondent’s own bargain and getting-expensive prices. Calibrated into probabilities, the answers produce a trial curve and a revenue index. It was presented at the American Marketing Association’s Advanced Research Techniques Forum in 1993.

Should I Use Van Westendorp or Gabor-Granger?

Choosing between Van Westendorp and Gabor-Granger depends on whether you already have candidate prices. Van Westendorp generally suits a familiar product with no known plausible price range, because respondents supply the prices. Gabor-Granger suits a shortlist of specific prices you’re choosing between. Many teams run the meter first and use its range to pick the Gabor-Granger test prices.

Does Van Westendorp Work for Business Software?

Van Westendorp typically works well for business software when the respondents are purchase decision-makers who know what comparable tools cost. Screen on purchase role rather than usage, and name the billing unit, such as per seat per month, in every question.

What Is a Good Van Westendorp Result?

A good Van Westendorp result has no external definition, because the price points depend on the description, audience and currency of the study. Compare against your own earlier waves with identical wording rather than against any published norm.

Which Curve Convention Should I Use?

Report Convention A, with the inverted not-a-bargain and not-expensive curves, as the primary acceptable range, and show Convention B alongside it. The two can differ substantially, and naming the one you used is what makes the result reproducible.

Is the Van Westendorp Method Reliable?

The Van Westendorp method is widely used for bounding a price range, though the validity evidence is thin and mixed, and it’s unreliable for predicting demand. One direct comparison, Kloss and Kunter in 2016, found its results biased but comparable to an incentive-aligned mechanism.

Broader research on direct questions finds hypothetical answers can overstate willingness to pay, so pair the meter with the extension, Gabor-Granger or a live test before committing revenue.

What Are the Limitations of the Van Westendorp Method?

The Van Westendorp method measures price perception instead of purchase, prices one product in isolation from competitors, and can’t value trade-offs between features or tiers. Its crossing points follow from a 1976 convention rather than economic theory, and its hypothetical answers can overstate willingness to pay. Gabor-Granger, conjoint analysis and in-market price tests cover each of those gaps.

What Drives the Cost of a Van Westendorp Study?

The cost of a Van Westendorp study is driven mainly by the number of cells and the cost of reaching screened respondents, since the instrument itself is short.

Each product or tier fielded as its own cell needs its own full sample, and screened business decision-makers on a panel generally cost more per complete than consumers. The meter is usually cheaper to design than conjoint, because it needs no experimental design.

Can I Run the Analysis in a Spreadsheet?

A spreadsheet can run the Van Westendorp analysis, using a cumulative count per price column and a lookup for each crossing. A spreadsheet makes the bootstrap intervals and the extension slow to build, which is why this guide provides the two AI prompts and why the R package pricesensitivitymeter exists.