Results from five price groups show that at ¥99, the “price × purchase intent” metric is highest; but when finance places a ¥35 per-user variable cost into the same data, the “(price − cost) × intent” metric is highest at ¥129. Neither calculation is wrong. The conflict arises from differing optimization targets, not from a need to re-run the survey.
Price research cannot automatically derive an “optimal price.” It can answer how the target audience perceives price, how stated purchase intent varies with price, and how they trade off price against features; the final price must also incorporate real conversion rates, costs, capacity, competition, channels, contracts, and strategic objectives. If it is not clarified what is being optimized before the study begins, no matter how sophisticated the model, unresolved business disagreements will simply be hidden in the decimal places.
First distinguish four questions—don’t mistake method names for objectives
| Decision to make | Appropriate research | Cannot directly answer |
|---|---|---|
| At what point users suspect quality is too low, and when it starts to feel expensive | Price perception and acceptable price range exploration | Profit-maximizing price, real demand volume |
| Acceptance of a fixed product at several candidate prices | Randomized unit price or sequential price purchase intent | Price after feature changes, actual conversion |
| Which features a package should include and how prices trade off | Conjoint analysis or discrete choice experiment | Sales volume not validated by model and market |
| How a real market price increase would affect purchase and retention | Transactional data, compliant field experiments, or phased policy evaluation | Causal effects when only attitude surveys are used |
The Van Westendorp price sensitivity meter often asks four monetary amounts—“cheap enough to doubt quality,” “a bargain,” “starting to get expensive,” and “too expensive to consider”—and uses intersections of cumulative curves to describe the price perception range. MIT OpenCourseWare pricing research materials list it alongside other methods. The intersections come from respondents’ price language, excluding unit costs, channel fees, or actual purchases, so they cannot be called a profit-optimal price.
Gabor–Granger–type methods ask purchase intent at multiple prices to form a price–response curve; showing sequential prices to the same person saves sample but reveals the research scope and introduces order and anchoring effects. Kamakura and van Wijngaarden’s research on this method also discusses psychological pricing patterns where lower prices can reduce intent in certain studies. When encountering non-monotonic curves, do not automatically smooth away “unattractive” points; first check whether lower prices are interpreted as lower quality, whether price order affected answers, and whether the sample size is too small.
The product being evaluated must be frozen enough to make a choice
“A user-friendly survey tool—how much would you pay?” Without billing units, feature boundaries, or alternatives, respondents each imagine their own different product. The price card should at minimum fix:
- Primary use scenarios and the results users can obtain;
- Core capabilities included and excluded, without treating wishes as validated performance;
- Whether billing is monthly, annual, per seat, or per response volume;
- Minimum purchase volume, contract term, taxes, service, and cancellation conditions;
- Current alternatives, plus a “no purchase / maintain status quo” option.
All price groups use the same product card and the same price framing. Do not write “less than ¥2 per day” for ¥59 while writing “¥129 per month” for ¥129; do not add premium support only on the high-price card. If significant misunderstandings still exist about the product concept, conduct a product concept test first; otherwise, price effects cannot be separated from information effects.
Who answers determines the result more than which price question is used
For consumer goods, the payer is usually close to the user, but for enterprise software there may be four roles: user, administrator, budget approver, and procurement. Users can judge feature value but may not know the budget; procurement knows contracts and alternative suppliers but may not experience daily friction. Mixing all four types into one average makes “willing to pay” lose a clear subject.
Sample eligibility should include recent category decisions or real needs, decision role within the organization/household, current alternatives, expected usage volume, and budget source. Surveying only existing high-value customers overestimates acceptance in the broader market; surveying only the general public without needs misreads reasonable non-purchase as price being too high. Reports should be limited to the population covered by the study, not extrapolated to un-surveyed roles and markets.
Five-group randomized unit price case: ¥99 has the highest revenue index
The following is a fictional SaaS monthly plan case, used only to illustrate calculations and not representative of actual questionnaire prices, conversion rates, or industry benchmarks. The study recruited 1000 people who had recently participated in purchasing similar tools and had budget recommendation or approval authority. Each person was randomly shown one price—¥59, ¥79, ¥99, ¥129, or ¥159—with 200 per group; product description, monthly payment terms, alternatives, and page were identical. The primary metric was the combined “very likely/likely to purchase” top-two-box proportion.
| Monthly price | High purchase intent | Price × intent | Contribution × intent (assuming ¥35 variable cost) |
|---|---|---|---|
| ¥59 | 116/200 (58%) | 34.22 | 13.92 |
| ¥79 | 104/200 (52%) | 41.08 | 22.88 |
| ¥99 | 90/200 (45%) | 44.55 | 28.80 |
| ¥129 | 64/200 (32%) | 41.28 | 30.08 |
| ¥159 | 42/200 (21%) | 33.39 | 26.04 |
The first index multiplies price by stated purchase proportion, so ¥99 is highest. If each new user incurs ¥35 per month in payment processing, service, and other variable costs, the second index uses (price − 35) × intent, making ¥129 highest. Costs are illustrative assumptions; replacing them with real payment fees, service hours, channel commissions, refunds, usage costs, and customer lifetime would change the ranking.
Both indices are “scenario indices per qualified respondent,” not revenue or profit forecasts. They do not include market size, actual conversion, customer acquisition cost, retention, capacity expansion, or competitive response. The correct use is to narrow the candidate range and expose objective conflicts: does the team prioritize early adoption, short-term unit contribution, capacity constraints, or long-term penetration?
A 13-percentage-point difference also needs uncertainty attached
The ¥99 group had 45% high intent; the ¥129 group, 32%—a point estimate difference of 13 percentage points. Using the large-sample proportion approximation for two independent randomized groups, the 95% difference interval is approximately 3.5 to 22.5 percentage points. The data support lower stated acceptance at ¥129, but the magnitude of the decline is still imprecise; it also does not tell us whether the contribution index will necessarily be higher in actual purchases.
Before the study, plan each price group sample size around the smallest meaningful difference; do not let a total sample of 1000 masquerade as 1000 per price. If many prices, roles, and packages are compared simultaneously, cherry-picking only the most favorable differences afterward amplifies chance findings. Primary comparisons, anomaly rules, and missing data handling can be pre-specified in an analysis plan.
Why not have everyone answer all five prices sequentially
Asking the same person price by price can estimate individual inflection points, but each subsequent answer necessarily knows the previous price. Respondents may guess the research intent, seek consistency, or anchor on the first price. Randomized starting points and ascending/descending prices can mitigate some order effects, but they cannot turn five answers into five independent buyers.
The randomized unit-price design has each person evaluate only one price, at the cost of needing independent samples for each price; the benefit is that price groups are closer to a single-variable comparison. Method choice depends on budget and decisions: early-stage scanning may use sequential prices, while validation of key candidate prices is better suited to randomized unit price. Results from the two should not be directly combined while ignoring design differences.
Package and feature trade-offs need discrete choice, not longer price questions
If the team actually needs to decide “which tier includes collaboration permissions, whether to bill per seat or per project, and how much support services are worth,” fixed-product unit-price acceptance cannot separate feature values. Discrete choice experiments let participants repeatedly choose among alternatives composed of common attributes at differing levels, with a status quo or no-choice option. The World Bank’s practical guide treats attribute selection, level setting, experimental design, questionnaire implementation, and econometric analysis as a continuous process; price coefficients can be used to estimate relative willingness to pay for other attributes.
More attributes are not necessarily better. First use interviews, transactions, and competitive information to identify the few attributes that truly enter the decision, exclude unrealistic or undeliverable combinations, then generate balanced and efficient choice sets. If the price range is too narrow, sensitivity cannot be estimated; too wide, and options become obvious. Repeated choices also require handling within-respondent correlation in analysis.
Why stated purchase intent in surveys cannot simply be multiplied by population to get sales
Surveys do not actually deduct money from accounts, nor do they require respondents to forgo alternative products. The NOAA expert report on hypothetical valuation notes that hypothetical markets may overestimate willingness to pay, and responses are affected by question framing and order; alternatives, payment vehicles, and budget constraints should be clearly presented, with conservative interpretation. The OECD on stated preference methods likewise notes that payment amounts in hypothetical choices can differ from real payments.
Therefore, 45% “likely to purchase” cannot be directly multiplied by the target population. Calibration can be strengthened step by step:
- Historical calibration using past “intent-to-actual purchase” relationships for similar products;
- Validation through reservations with explicit prices, post-trial upgrades, or sales quote results;
- Real market tests designed without deception, discrimination, and in compliance with contract and consumer rules;
- Ongoing observation of conversion, refunds, usage, support costs, and retention, rather than first purchase only.
Price experiments may involve customer fairness, published pricing, tax, channel, and contract implications; one cannot secretly offer different prices to identical conditions for the sake of “real behavior.” Specific plans should be jointly reviewed by product, finance, legal, and research teams.
A one-page conclusion that can genuinely enter a pricing meeting
Conclusions can be written as: “Under this sample with relevant purchasing experience and a fixed monthly plan, stated purchase intent generally declines from ¥59 to ¥159. The price × intent index is highest at ¥99; after adding an illustrative ¥35 per-user variable cost, the contribution index is highest at ¥129. Neither is a sales or profit forecast. Next steps include supplementing with real quotes/upgrade behavior near ¥99 and ¥129, and running scenario analysis on price, support costs, and retention.”
What cannot be written is: “The survey proves ¥129 is the optimal price.” The former puts decision goals, sample, scenario, calculation, and evidence gaps all on the table; the latter lets a hypothetical choice substitute for a business decision.
Wenjuanpai can be used to screen qualifying roles, randomly display price cards, set no-purchase options, and collect choice reasons; the experimental design, estimation, cost modeling, and final pricing for complex discrete choices still require appropriate specialists. The tool is responsible for stably executing study designs; it cannot choose among profit, growth, or fairness trade-offs for the team.
Price perception, stated preference, and the boundary of real payment
- MIT OpenCourseWare — Measuring Customer Reactions to Prices
- Kamakura & van Wijngaarden — A model for the effects of psychological pricing in Gabor–Granger price studies
- World Bank — How to Conduct a Discrete Choice Experiment for Health Workforce Recruitment and Retention in Remote and Rural Areas
- NOAA — Report of the NOAA Panel on Contingent Valuation
- OECD — Theory and methods of mortality risk valuation