Running a Van Westendorp Study on an Online Panel
Learn why rigorous screening and data cleaning matter more than the method itself.
Running a Van Westendorp study on an online panel is straightforward. Getting defensible numbers takes more care. The method can identify a credible price range, but only when the right buyers assess the same offer, answer in the right sequence, and pass logical checks. That range is a starting hypothesis.
What the Van Westendorp Price Sensitivity Meter actually measures — and what it cannot
The Van Westendorp method measures how people see prices but does not measure demand. Its four questions locate different thresholds, from a price so low that quality seems doubtful to one so high that buying is out of the question. Between them sit “bargain,” where the offer feels like good value, and “getting expensive,” where hesitation begins but purchase remains possible. Respondents type their own numbers into open numeric fields. That avoids anchoring them to a researcher’s price list.
The resulting curves identify an acceptable range and points where different perceptions balance. They do not explain why respondents chose those thresholds or how many units would sell. The name “Optimal Price Point” is particularly easy to overread. It balances two kinds of rejection. This is the right tool for scoping a range, with Gabor-Granger providing a later step when a specific price is needed.
Why the same method produces trustworthy ranges on one panel and garbage on another
Van Westendorp cannot tell which respondents have never bought in the category. Someone who would never buy in the category can compress the acceptable range, while someone with the right job title may never have approved a relevant purchase. A panel of “marketing managers” is not necessarily a panel of buyers.
Existing customers and prospects approach the offer differently, so combining them can describe neither group accurately when the question distinguishes acquisition from retention pricing. Product familiarity creates another trap. A participant assessing a SaaS tool may price the incumbent they know. The range holds only if participants are pricing the intended product.
Screener design and sample size decisions that make the curves defensible
To avoid those failures, screen for whether people take part in purchases as well as whether they fit the demographics. Require involvement in a similar purchase within the past year or active evaluation now. Qualify economic buyers, technical evaluators, or strong influencers whose company size, industry, or use case matches the target segment. Ask about at least one competitor or substitute. Route people who would never buy in the category to a follow-up about why, but exclude them from pricing analysis.
Establish the reference frame before asking for thresholds. Ask about participants’ current purchases and prices paid, including comparable SaaS platforms and the features they use. For B2B, confirm whether they approved, vetoed, or built the business case for a relevant purchase within the past year. Keep customers and prospects in separate sample cells. Modern online panels with verified demographic and behavioral filters make this feasible without long recruitment timelines.
Plan for roughly 150 to 200 usable completes after cleaning for a single segment. Below that, intersections can shift materially with each response, and fewer than roughly 200 usable responses leaves crossing-price estimates difficult to defend. Segmented analysis needs substantial numbers within each segment. Geography needs separate treatment too. Recruit national or regional cells and compare purchasing-power-parity-adjusted ranges rather than nominal prices across markets with different income levels.
Writing the product stimulus and sequencing the four questions correctly
Give every respondent the same concrete offer to price. For SaaS, state which plan tier is on offer, the seat-count basis, and which core features it has. For physical products, specify size, materials, and box contents. Keep that description in view while people answer the four questions. Leave competitor prices, promotional anchors, and percentage discounts out of it, because those figures steer the answers.
Ask in this fixed order: too cheap, bargain, getting expensive, and too expensive. First ask when the price is so low that quality becomes doubtful, then when it feels like a great buy. Next ask when it starts getting expensive but remains worth considering, then when it becomes too expensive to consider. Conveo’s PSM guide warns that changing this order introduces anchoring effects.
Keep all four answers as open numeric fields. Include a skip or “not applicable” option for people who genuinely would never buy, and remove them from pricing analysis. When comparing configurations, use monadic testing with one configuration per respondent and the sample split across cells. Otherwise, earlier configurations anchor later answers. In each national cell, use the local currency and show it in a format people know.
Validating responses before they reach the analysis — the data cleaning step
Validation is mandatory. For a response to be usable, its prices must follow the order Too Expensive ≥ Getting Expensive ≥ Cheap (Bargain) ≥ Too Cheap. A “too cheap” answer above “too expensive” is impossible and cannot be repaired through imputation. Prefer instrument-level checks over judgment calls after fieldwork. For equal adjacent thresholds, document the rule and apply it consistently.
Logical checks do not replace fraud controls. Standard attention checks catch inattentive respondents, sophisticated fraud escapes those checks, and platform-level identity and behavioral verification reduces the risk while leaving some risk in place. Choose panel providers with that distinction in mind. Remove invalid responses instead of averaging or imputing them, then recount each cell. Set recruitment targets with cleaning attrition in mind so the remaining usable completes still meet the sample floor.
Plotting the curves and reading the four intersection points without over-interpreting them
As price rises, the “too cheap” and “bargain” curves fall because fewer respondents see suspiciously low pricing or a great deal. The “getting expensive” and “too expensive” curves rise as hesitation and rejection increase. Those movements describe perceptions across the screened sample. They are not a record of purchases. Each intersection therefore supports a narrower claim than a pricing meeting may want to attach to it.
The Point of Marginal Cheapness, or PMC, marks the lower bound of the acceptable range, where “too cheap” crosses “not cheap,” the inverse of bargain. The Point of Marginal Expensiveness, or PME, marks the upper bound. A price below the lower bound can make the product seem poor in quality. The upper bound marks increasing resistance. Neither boundary shows that demand stays the same between them.
The Optimal Price Point, or OPP, is where “too cheap” and “too expensive” cross, balancing rejection on either side. The Indifference Price Point, or IPP, is where bargain and getting expensive cross, with equal shares seeing a deal or a stretch. The latter can serve as a market-implied reference price. Neither is a revenue-maximizing result. Hypothetical willingness to pay overstates real-world acceptance, so the range remains a hypothesis about screened buyers.
Adding Newton-Miller-Smith purchase-intent questions to estimate reach
The Newton-Miller-Smith extension adds two purchase-likelihood questions, usually at the prices each respondent gave as acceptably cheap and acceptably expensive. Respondents rate how likely they are to buy at those amounts. This adds a demand dimension within the same fieldwork session. It is useful when finance needs volume or revenue assumptions, or when a wide range leaves the team choosing between OPP and IPP as a starting point.
The extra questions do not establish actual buying behavior. Their answers are still stated intentions, with the same risk of hypothetical overstatement. Treat the extension as a way to rank price points by likely reach. It improves the direction of the estimate. A decision requiring a sharper price recommendation calls for a further test.
Following Van Westendorp with Gabor-Granger to sharpen a range into a specific number
Gabor-Granger tests purchase intent at discrete prices drawn from the acceptable range. Each respondent sees a single price from a predetermined list and answers a direct purchase-intent question, using either a five-point likelihood scale or yes/no. Aggregating responses across prices produces a step-wise demand curve. Price multiplied by estimated reach identifies the revenue-optimizing candidate. It remains a candidate based on estimated demand.
The order gives the two methods distinct jobs. Van Westendorp establishes useful territory, while Gabor-Granger tests specific prices within it. Without the first step, a predefined list can land outside the acceptable zone. Without the second, a range does not settle the launch price.
For early concept validation, category entry without a reference price, or budget-constrained scoping, Van Westendorp alone can be sufficient. For established-product launches, pricing pages, or revenue-maximizing recommendations, run both in sequence. When direct substitutes make the question about which features justify a higher tier, use conjoint analysis. Van Westendorp can still establish the price range that conjoint should test.
How AI-moderated interviews and synthetic panels change the execution calculus
AI-moderated interviews can help explain why people chose each threshold. A follow-up asking what made a price too expensive connects the number to the respondent’s reasoning. Probing and synthesis can add that qualitative layer at scale without adding analyst time. The economics have changed as well. Lower cost per complete and end-to-end platforms reduce the need to coordinate separate panel, programming, and analysis work.
Synthetic panels change speed and cost, but not the need for human validation. Qualtrics and Toluna offer synthetic respondent capabilities with human panels, enabling faster and cheaper fieldwork. For decisions going to a board, investor, or public pricing page, human-only panels remain the most defensible choice. Hybrid studies suit iterative development when synthetic scoping can be followed by a human-validation cell. Synthetic-only work belongs in directional hypothesis generation for familiar categories, never as the final input to a new offering’s launch price.
AI-native platforms also combine verified human panels with synthetic modeling, shortening the path from a pricing question to a usable range from weeks to hours. That speed does not remove the design requirements. Screening, a shared product reference frame, and response validation still determine what the result supports. For repeated research across markets, tiers, or versions, a foundational behavioral model of the target audience offers a more cost-effective base than commissioning a fresh agency study each time.
The decision checklist: what to verify before, during, and after fieldwork
Before fieldwork, check that the stimulus defines one specific offer and the screener establishes recent purchase involvement, authority, and category experience. Keep customer and prospect cells separate, and include a warm-up about current products and spending. Budget enough completes to absorb cleaning losses in each segment. Evaluate the provider’s identity checks and fraud controls. For cross-market work, plan separate cells and purchasing-power-parity adjustments.
During fieldwork, keep the questions in the fixed order from too cheap through too expensive. Keep answers open and numeric, with no predefined price list. Apply instrument-level logic checks before submission. Show each respondent only one configuration when testing alternatives. Match currency displays to the local market.
After fieldwork, remove invalid price ladders and apply the documented rule for equal adjacent thresholds consistently. Recount usable completes before calculating curves for each segment. Look at the results for customers and prospects separately before deciding whether to merge them. Combining customers and prospects requires both substantially overlapping ranges and justification from the pricing decision.
At interpretation, distinguish rejection balance from revenue optimization and price credibility from demand. Treat the upper bound as resistance and recognize that a range may be inflated without competitive framing when direct substitutes exist. Use Gabor-Granger when the launch decision needs a number, or conjoint when it needs feature-price trade-offs. Build toward a reusable audience model for repeated research. Well-screened, validated studies followed by the right next test yield more than a one-time exercise.
