How to Run: Pricing Research Study

Introduction

To run a pricing research study, choose the method from the pricing question. Use Van Westendorp when you don’t know what range is plausible, Gabor-Granger or a monadic test when you have three to six candidate prices, and conjoint when price trades against features. Field the study to people who could buy, size every price cell from the precision you need, and treat the result as a ranking of prices, not a revenue forecast.

Last updated October 2026.

What a Pricing Research Study Is

A pricing research study measures how the people who could buy a product respond to its price. Teams use one to choose a price, a price range, or a price structure before committing to it in market. It isn’t a single method. It’s a family of survey instruments, and each answers a different question.

This guide owns the choices that come before the analysis: which method fits the decision, who to ask, how many people each price needs, and what the result can and can’t tell you. The companion Van Westendorp guide runs the Price Sensitivity Meter analysis line by line. The Gabor-Granger guide and the conjoint guides do the same for their methods.

The Five Instruments in Circulation

Five instruments come up in nearly every pricing discussion:

Van Westendorp Price Sensitivity Meter
Gabor-Granger
Monadic price test
Conjoint with a price attribute
In-market price test

The first four are surveys. The fifth is an experiment on real buyers, and it’s the standard the other four are judged against.

Direct and Indirect Methods

Breidert, Hahsler, and Reutterer (2006) sort survey methods into direct and indirect in "A Review of Methods for Measuring Willingness-to-Pay," published in Innovative Marketing 2(4), pages 8 to 32. Direct methods ask about price outright. Indirect methods infer price sensitivity from choices that include price among other attributes.

Van Westendorp, Gabor-Granger, and monadic tests are direct. Conjoint is indirect. The split matters later, because the evidence on which kind overstates more runs against what most teams assume.

Price Level and Price Structure

Teams often merge two questions. Price level is how much to charge. Price structure is what to charge for, such as per seat, per usage unit, a flat fee, or a tiered bundle. The methods here measure level for a structure you’ve already chosen.

To compare structures, describe each as its own product and price them separately, or use conjoint with structure as an attribute.

When to Run a Pricing Research Study

Run a pricing study when a price decision is live and reversing it would be expensive. The usual triggers are a launch with no reference price, a new tier or package, a price increase on an existing base, and entry into a market where your price has never been tested.

Three Conditions That Make a Survey the Right Instrument

First, you can’t run an in-market test. A team with a checkout it can split should usually test real prices on real traffic, because the evidence is stronger and often cheaper to get.

Second, you can name the decision. "Which of these four prices should we launch at" is a study. "Learn about price perception" is a chart in search of a meeting.

Third, you can reach people who could plausibly buy. A study fielded outside the category measures someone else’s wallet.

Before commissioning anything, write a one-paragraph decision brief. It names the decision, the range of prices the business would actually defend, the smallest difference in acceptance that would change the choice, and the segments the result has to cover. If the team can’t write that paragraph, the study isn’t ready, and no method will rescue it.

Where Survey Pricing Research Earns Its Cost

Surveys fit early decisions where no market exists yet to test in. They also fit business products whose sales cycles make live experiments slow, and they work well for narrowing a wide range to a shortlist before any in-market test runs.

Signals That a Pricing Study Is Premature

Some studies get commissioned before the product can be priced. Three signals give it away: a product description that changes week to week, a packaging decision nobody has made, and a target buyer the team can’t name yet. Each one makes the result describe a product that won’t ship. Settle the package and the buyer first. A curve for the wrong product costs more than the delay.

When Not to Run a Pricing Research Study

A pricing survey can’t answer five questions. Each one below names the instrument that can.

It Can’t Tell You What People Will Actually Pay

Every survey method reads a hypothetical answer. A respondent who calls a price acceptable has spent nothing, and the gap between that answer and a purchase is the subject of the critique section below.

The better instrument is an in-market price test: an A/B test on a live checkout, or a staged rollout by region or segment. Where a survey is unavoidable, the Becker-DeGroot-Marschak procedure asks respondents to commit to a real purchase at a randomly drawn price. Wertenbroch and Skiera (2002) describe it at the point of purchase in the Journal of Marketing Research 39(2), pages 228 to 241.

It Can’t Tell You How Price Trades Against Features and Competitors

Van Westendorp and Gabor-Granger price one product in a vacuum. A respondent who hears your price rarely hears the price of the alternative they’d otherwise buy, and real price sensitivity often lives in that comparison.

The better instrument is conjoint analysis with a price attribute and a none option, which forces respondents to trade price against what it buys.

It Can’t Tell You How Existing Customers Will React to a Price Increase

Current customers are anchored on what they pay today. Asked about a higher price, they report outrage they won’t act on or acceptance they’ll reverse at renewal. Stated churn intent isn’t churn.

The better instrument is a staged rollout or a holdout group on the affected cohort, read against renewal and churn after the change ships. A survey of that cohort can add reasons, but it can’t add the behavior. This limit changes what teams do more than any other here, because a price increase is the decision most often researched with the wrong population.

It Can’t Tell You Which Features Belong in Which Tier

A pricing study prices a package. It doesn’t say what the package should contain, and a tier built around the wrong features tests poorly at any price.

The better instrument is MaxDiff, which ranks a long feature list by relative importance so the package you price reflects what buyers value.

Start with How to Run a MaxDiff Study and run the analysis with MaxDiff Analysis with Claude. TURF analysis, which finds the feature combination that reaches the most buyers, is the natural next step, and no guide covers it yet.

It Can’t Tell You Why a Price Feels Wrong

Every curve in a pricing study is a quantity with no reasoning attached. A chart shows where "too expensive" climbs. It never says whether the objection is budget, perceived value, or a competitor’s anchor.

The better instrument is qualitative interviews with people who rejected the price. A rejected price can mean the buyer lacks budget, doubts the value, or has a cheaper option in mind, and each calls for a different response.

Run the interviews separately and let each method do its own job. When to Run AI-Moderated Interviews explains where interviews fit next to surveys.

Study Design and the Fork That Decides Everything Else

Choose the method from the pricing question in front of you. Don’t default to whichever template the team already has.

The Pricing Method Decision Tree

Work down this tree and stop at the first branch that matches:

Can you run a live price test on real buyers? If yes, run it, and use a survey only to narrow the candidates first.
Do you already know a plausible range? If not, run Van Westendorp to find one.
Does price trade against features, tiers, or named competitors in this decision? If yes, run conjoint with a price attribute.
Do you need each candidate read without anchoring from the others? If yes, run a monadic test, one price per respondent.
Otherwise, run Gabor-Granger on three to six candidate prices inside the range.

The tree is also a sequence. Teams with an unknown range usually run Van Westendorp first, then Gabor-Granger or a monadic test on the shortlist, then an in-market test for the final call.

The Fork in One Table

The five instruments differ in what they return and what they cost:

The method follows the question, and cost rises as the evidence gets closer to a purchase.

What Each Instrument Does

Van Westendorp (1976) asks four questions that have respondents name prices rather than react to them: too cheap, a bargain, getting expensive, too expensive. The source is Proceedings of the 29th ESOMAR Congress, pages 139 to 167. It suits products with no reference price, because respondents anchor on their own sense of value instead of a number you supplied.

Newton, Miller, and Smith (1993) added two purchase-likelihood questions to turn the range into trial and revenue curves, presented at the American Marketing Association’s Advanced Research Techniques Forum.

Gabor-Granger shows a specific price and asks whether the respondent would buy, then moves up or down a ladder. A monadic test shows each respondent exactly one price, so no answer is shaped by an earlier price. Conjoint shows complete configurations with price as one attribute.

The Recommendation

Run direct methods first and indirect methods only where the trade-off is the question. Schmidt and Bijmolt (2020) found indirect methods overstated willingness to pay more than direct ones, which cuts against the assumption that conjoint is the safer instrument. That doesn’t make conjoint wrong. It makes it the tool for trade-off questions rather than a default for every price decision.

A Worked Three-Study Sequence

Here’s an illustrative sequence with invented numbers and no currency. A team launching a per-seat product has no reference price, so it runs three studies in order.

Study one is Van Westendorp with 400 valid respondents. It returns an acceptable range of 9 to 21 per seat per month. Study two is a four-cell monadic test at 10, 13, 16, and 19, sized for an eight-point gap, which needs about 2,424 completes. Acceptance separates 19 from the rest but not 13 from 16.

Study three is an in-market test. The team splits new sign-ups between 13 and 16 and reads trial conversion and 90-day retention. The surveys narrowed a wide range to two prices, and real behavior made the final call.

Sprig Note: What the Platform Can’t Build. Sprig documents no pricing question type, so there’s no Van Westendorp, Gabor-Granger, or price-ladder type among its 15 question types. Its survey logic documentation describes randomizing the order of questions, pages, and answer options, and documents no way to assign respondents randomly to conditions, so a true monadic test can’t be built in-product and the sample frame has to be split before fielding.

The same documentation says randomization and response piping can’t be used at the same time. These limits are as documented in October 2026, so confirm current behavior with your account team. Conjoint is a native question type with an optional "None of these" option, but it’s available to Enterprise teams on link surveys only.

Writing the Instrument

The instrument is where pricing studies quietly fail. Lift the specification below and change only the product description and the price ladder.

The Product Description Comes First

Every price answer is given against whatever the respondent believes the product is. Show a short, concrete description before any price question, naming what the buyer gets, the billing unit, and the term.

"A scheduling tool for teams" invites a different price than "unlimited shared calendars for up to 50 people, billed per seat per month." The second is a product a respondent can price.

Test the description on someone outside the team. If they can’t say what they’d get for the price, a respondent won’t be able to either, and the study will measure confusion rather than value. Keep the same description, word for word, in every cell and every wave.

The Price Ladder

Space three to six prices across the range the business would actually charge. Set them at points a buyer would recognize, including any competitor reference price, instead of even increments. A ladder built from arbitrary steps tests distinctions nobody makes.

Three anchors belong on almost every ladder: the price you charge today if one exists, the nearest competitor’s reference price, and the top of the range the business would defend. Keep the steps wide enough that a buyer would notice them. A ladder of four prices that differ by a few percent each can’t separate in any realistic sample.

Keep the ladder inside the range a Van Westendorp study returns when you’ve run one. Whatever you choose, write the ladder into the study plan before fielding.

The Instrument Spec

| Instrument | Question wording | Scale | Notes | |:---------------:|:------------------------------------------------------------------------------------------------:|:--------------------------------------------:|:-----------------------------------------------------------------------------:| | Gabor-Granger | If this product were available at [PRICE] per [BILLING UNIT], how likely would you be to buy it? | Five points, very unlikely to very likely | Top two boxes count as acceptance. Randomize ladder order in a fixed sequence | | Monadic test | The same question, shown once, at the one price assigned to that respondent | Five points | One price per respondent, cells launched together | | Van Westendorp | Four numeric-entry questions: too cheap, a bargain, getting expensive, too expensive | Open numeric entry in the billing unit shown | Full wording and analysis live in the companion Van Westendorp guide | | Open-text probe | What would need to be true for this product to be worth [PRICE] to you? | Open text | Asked only of respondents who rejected the price, placed last |

For a branching Gabor-Granger design, start at the middle price. Move up one rung after a top-two answer and down one rung after anything else, and stop when the answer changes direction. Each respondent answers only some rungs, so the base at each price has to be counted after fielding rather than assumed.

Wording Traps That Move Every Answer

Four details can move a price answer:

State the currency, since respondents in a multi-currency market will otherwise guess
State whether the price includes tax, which often differs by market
State the billing unit and the term together, such as per seat per month billed annually
Keep the unit identical across every question, because a respondent who switches from monthly to annual midway produces a record that looks real and isn’t

A monthly and an annual figure for the same product are often answered as if they were different products.

Sprig Note: Building This Instrument. Validated text supports a numeric format, which captures the Van Westendorp prices. Response piping carries a respondent’s own answer into later questions, but Sprig’s survey logic documentation says randomization and piping can’t be used together, so a survey that pipes each respondent’s own answer can’t also randomize question order.

The Design Agent can draft a study from a written brief with logic in place. Sprig publishes no validation that it programs a branching ladder correctly, so preview every branch path by hand.

Scale and Scoring Choices

Scoring decisions made at design time decide what the analysis can say. Make them before fielding and write them into the study plan.

Acceptance in Gabor-Granger and Monadic Tests

Count the top two boxes of a five-point purchase-likelihood scale as acceptance, which is the common convention rather than a validated rule. A top-box-only count is a legitimate stricter rule for high-consideration purchases. Choose one and report it. Switching definitions after seeing results is a way of choosing the answer, so state the rule in the plan.

Use the same five-point scale in every pricing question across every cell. A seven-point scale in one wave and a five-point scale in the next makes the acceptance shares incomparable, and the difference gets read as a market shift. If you have to change the scale, treat the second study as a new study and don’t compare it to the first.

Open Numeric Entry or a Price List

Van Westendorp is designed for open numeric entry, where the respondent types any price. A dropdown of preset prices is faster to answer, but it imposes your grid and caps the range at its ends. A respondent whose true ceiling sits above your highest option gets recorded at it.

Keep entry open unless respondents generally can’t name a price at all. That usually means the product description needs work.

Round-Number Heaping

Respondents answer in round numbers, so open-entry data clusters at values like 10, 15, 20, and 25. The curves become step-shaped and intersections land near those values. That’s a property of the data, not of the market.

Purchase Probabilities in the Extension

The Newton-Miller-Smith extension converts each likelihood answer to a probability. Those weights are a researcher assumption, not a published constant. Whatever weights you choose, print them next to the result. Two analysts using different weights on the same data will report different optimal prices, and both are internally correct.

Sample Size and Precision

Size a pricing study from the margin of error you can tolerate at each price, not from a circulated threshold. SurveyMonkey’s comparison of Gabor-Granger and Van Westendorp says that surveying "at least one hundred respondents that fit the profile of your customers" gives a useful read, with no derivation behind the figure.

Commonly circulated floors of 100 to 200 per segment trace to no calculation we could find.

So the derivation below is the asset. Every figure assumes a 95 percent confidence level, equal cells, and no design effect. The comparison table uses the unpooled-variance formula for two proportions.

The Margin of Error at One Price

Acceptance at a single price is a proportion. Its margin of error is 1.96 × √(p(1−p)/n), where p is the acceptance share and n is the number of respondents answering at that price.

The margin is widest at 50 percent acceptance, so plan from that column unless a prior study shows acceptance well away from the middle. At 100 respondents per price, 42 percent against 50 percent acceptance is a tie, and calling the higher one the winner reports noise.

How Many Respondents Each Price Needs

Comparing two prices is a harder test than reading one. The table below shows respondents needed per price to detect a true gap at 95 percent confidence and 80 percent power, with the second price’s acceptance set that many points above the baseline. Values are rounded up.

| Baseline acceptance | 3-point gap | 5-point gap | 8-point gap | 10-point gap | |:-------------------:|:-----------:|:-----------:|:-----------:|:------------:| | 20% | 2,940 | 1,092 | 444 | 291 | | 30% | 3,760 | 1,374 | 547 | 354 | | 50% | 4,353 | 1,562 | 606 | 385 |

A four-price monadic test that must detect an eight-point gap at a 50 percent base needs about 606 respondents per price, or about 2,424 completes. Instead of asking how many respondents are enough, ask how small a gap would change the decision. That trade is the real output of sample planning.

What Multiplies the Requirement

A monadic test needs the full n in every cell, so four prices at 200 each is 800 completes. A fixed-sequence Gabor-Granger asks every respondent every price, so 200 respondents give 200 answers at each rung, bought with anchoring. A branching ladder leaves the end prices thin.

Segments multiply everything. Every segment you plan to report needs the full requirement on its own, so decide the reporting cuts before fielding.

Van Westendorp intersections have no closed-form margin of error, because each point is a crossing between two empirical curves. The companion guide handles that with a bootstrap. The same margin arithmetic underlies the margin-of-error table in the NPS guide, which applies it to a different measure.

Minimum Base for Any Reported Cell

Report only cells whose base gives a margin you’d accept for the decision, using the first table above, and treat a base of zero as the extreme case rather than a missing row. Write the minimum into the analysis plan before fielding.

A pilot makes the plan better. Its acceptance share is a realistic value for p in the formula, and when acceptance sits well away from 50 percent the required sample shrinks. At a 20 percent baseline, detecting an eight-point gap takes 444 per price instead of 606. Run the pilot, read the acceptance, and then size the full study.

Audience, Targeting, and Screening

A study answered by non-buyers prices someone else’s wallet. The acceptance a non-buyer reports isn’t the acceptance that produces revenue.

Prospects and Customers Are Different Studies

Decide before fielding whether you’re pricing for people who’ve never bought or for people who already pay you. They answer differently, and mixing them produces a curve that describes neither.

Prospects are the right sample for a launch or a new market. Existing customers are the right sample for a change to something they already use, with the anchoring caveat from the earlier section held in view.

Who Belongs in a Prospect Sample

Restrict the sample to people who could make or influence the purchase. For business software that means a role screen and a company-size screen. For a consumer product it means a category-purchase screen.

The person who uses a product and the person who approves the spend often answer differently. Decide which one the price depends on and screen for that role. Screen on behavior, not interest: for example, "Have you paid for a scheduling tool in the last 12 months" beats "Would you be interested in a scheduling tool," which tells respondents what you hope to hear.

Keep Price Out of the Screener

A screener that mentions price, budget, or plan names primes every answer that follows. Keep it about category behavior and role, so the first price a respondent sees is the one you’re measuring.

Who Belongs in a Customer Sample

Target the customers the change affects, usually by plan and tenure. A long-tenured customer on a legacy plan and a new customer on the current plan have different reference prices. Survey the cohort that will receive the change, ideally inside the product at the moment they’re thinking about it.

Sprig Note: Reaching Each Population. For prospects, research panels offer 300+ targeting attributes across demographic, professional, behavioral, and firmographic criteria. For customers, in-product surveys on web and in mobile apps can target a plan or an upgrade page. Two limits apply.

Sprig’s own pages publish different participant counts, so confirm reach for your audience with your account team, and a respondent reached on an upgrade page is already anchored on the price shown there.

Fielding and Delivery

Put the price block early and keep the survey short. A respondent who reaches the price questions after a long block of unrelated items answers them faster and less carefully.

Split the Sample Frame for a Monadic Test

A monadic test assigns each respondent one price at random. Where the survey tool doesn’t document random assignment to conditions, randomize yourself before fielding: split the contact list into as many equal groups as there are prices, or launch one identical study per price to the same audience at the same time.

Launch the cells together so time of day and day of week fall evenly across them. Two audiences assembled at different times aren’t comparable.

After fielding, compare the cells on the screener variables before you read any acceptance share. If one price cell skews toward larger companies or longer-tenured customers, the gap between cells partly reflects who landed there, and the comparison needs a caveat or a reweighting.

Order Effects in a Fixed-Sequence Ladder

A respondent who sees a low price first reads later prices as expensive, and one who sees a high price first reads them as bargains. Randomize the ladder order in a fixed-sequence design, or accept that ascending order anchors low and report it.

Choose the Channel From the Population

An emailed link reaches your most engaged customers first. A panel reaches people with no relationship to your product. The same pricing survey fielded both ways returns different curves, so pick the channel from the population you named, and read How to Run a Product Concept Test if the earlier question is whether anyone wants the product.

Pilot Before the Full Launch

Field a small pilot first, enough to see every branch of the survey several times, and read the records one at a time. A numeric question with no validation, a branch that loops, or a piped price that shows blank is often invisible in the editor and obvious in the data.

Invitations and Incentives

Keep price out of the invitation as well as the screener. An invitation that promises a discount or names a plan anchors every respondent before the first question. Keep incentives modest and identical across cells, since a large reward attracts people motivated by the reward and makes a monadic comparison unfair.

Sprig Note: Fielding Controls. Response quotas can balance a sample across segments, and randomization is documented for answer options, questions, and pages. Quotas are available only to Enterprise teams and count completed responses. Random assignment of a respondent to one of several price conditions isn’t documented, which is why the split-frame approach above is the recommended route.

Timing and Cadence

Run pricing research before a launch, a packaging change, or a price increase, early enough to change the plan. Re-running it on a schedule produces a line that moves for reasons nobody can attribute.

Run a study before a launch, before a packaging change, and before a price increase, early enough that the result can still change the plan. A study that lands after the price is announced is a post-mortem.

Re-run when the market has moved enough that your reference points are stale: a competitor repricing, a new entrant, or a material change to what your product includes. Hold the product description, question wording, and price ladder fixed. Changing any of them turns the second study into a new study.

Time a customer study around the customer’s calendar, not yours. A survey about a renewal price is read most honestly near the renewal, when the reference point is concrete and the respondent has a real decision in mind.

Allow time for the full sequence when the range is unknown. A range study, a shortlist test, and an in-market check are three pieces of fieldwork, and compressing them into one survey reintroduces the anchoring each step was designed to avoid.

Quality Control and Data Hygiene

Pricing data hides bad responses well, because a careless answer is still a number. Write the exclusion rules into the plan before any data arrives.

Consistency and Outliers

A valid Van Westendorp respondent names prices in order: too cheap at or below a bargain, a bargain at or below getting expensive, and getting expensive at or below too expensive. Ties are common because of heaping, so keep them and report how many there were. Exclude respondents whose prices break the order, and report how many you excluded. A high exclusion rate usually means the questions were unclear.

One inconsistency is common enough to check on its own: a bargain price entered above the getting-expensive price. It usually means the respondent read the two middle questions in the wrong order, which is a wording problem rather than a respondent problem. Fix it in the pilot, since no exclusion rule can repair it after fielding.

Numeric entry also invites extremes. Set a rule in advance, such as excluding answers beyond a stated multiple of the column median, and apply it the same way everywhere.

Speed and Straight-Lining

Compute the median completion time and read the fastest few percent of records before deciding what to do with them. In a Gabor-Granger ladder, a respondent who gives the same answer at every price is either indifferent to price or not reading, and the record usually says which.

What to Do With Flagged Records

Apply the written rule, report the counts, and keep excluded records in a separate file. A rule applied after seeing who produced an inconvenient answer isn’t quality control.

Analysis: The Core Calculation

Each method has one core calculation, and a spreadsheet can run all of them. The difficulty is in the definitions and exclusions, not the arithmetic.

Gabor-Granger acceptance at each price is the share answering in the top two boxes. The revenue index is price times acceptance, and the highest index is the leading candidate. Take an illustrative ladder of 10, 13, 16, and 19 with acceptance of 62, 51, 43, and 28 percent. The indices are 6.2, 6.6, 6.9, and 5.3.

With 400 respondents per price, the gap between 13 and 16 is smaller than its sampling error, so the two are tied and 19 is behind.

Monadic acceptance is a simple proportion in each cell, compared across cells with a two-proportion test.

Before computing anything, apply the exclusion rules from the plan and write down the base at each price. A revenue index calculated on a base of 40 at the top of a branching ladder looks as precise as one calculated on 400, and it isn’t. Report the base next to every share.

The companion Van Westendorp guide covers the curves and intersections, and the Gabor-Granger and conjoint analysis guides cover the rest.

Benchmarks and What a Good Result Looks Like

There’s no cross-industry benchmark for an acceptable price range, an optimal price point, or a Gabor-Granger acceptance rate, and there can’t be one: every one of those outputs is denominated in the category, currency, and reference prices your respondents brought into the survey, so a competitor’s optimal price point describes their buyers, not yours.

Reproducing a vendor’s table of typical acceptance rates imports someone else’s product, audience, and ladder into your decision.

What to Compare Against Instead

Compare against your own prior studies on the same instrument, with the same product description, fielded to the same population. That’s the only comparison where a change in the curve means a change in the market.

If you have no prior study, your first study is the baseline. Record the product description, the ladder, the audience, and the dates, because the second study can only be compared to the first if you can reproduce its conditions.

Compare candidate prices within the study, which every method here does well. And compare against behavior where you have it: a price you already charge has a real conversion rate, and a survey that reads your current price as far outside the acceptable range is telling you something about the survey.

What a Good Result Looks Like

A good range study has an acceptable range narrow enough to change the decision and an exclusion rate within the limit you wrote into the plan. A good Gabor-Granger result has a revenue peak whose interval doesn’t overlap its neighbors. A flat result isn’t a failed study. A curve that barely falls across the ladder usually means the ladder was too narrow.

Interpreting and Acting on the Result

Read the output as a comparison between candidate prices under stated preference. Present the range, the candidates inside it, and how confident the study is about each, not a single "optimal" price.

Reading a Revenue Curve

A revenue index that peaks sharply names a clear leading price. One that runs flat across two or three prices means the study can’t separate them. That’s a real finding, and the business can choose among them on strategy.

When the Result Points at Structure

Sometimes the output says more about structure than level. A very wide acceptable range, a split between segments, or a revenue curve with two humps means different buyers value the product differently, and a single price will leave money on the table with one group or lose the other. That’s an argument for tiers. Design a package for each group and price them separately.

Segment Before You Conclude

Cut the result by the segments you planned, provided each clears the base you set. The cross-tab guide covers comparing a measure across segments in an AI client.

What to Hand the Decision-Maker

Give them the range, two or three candidate prices, the interval around each estimate, the exclusion count, and one sentence stating that these are stated-preference results. Then name the in-market test that would confirm the choice.

A Worked Read

Take an illustrative study with 400 valid respondents. The acceptable range runs from 9 to 21 per month. A follow-up test at 13, 16, and 19, also with invented figures, puts stated revenue at its peak at 16, with 13 inside the noise.

The obvious reading is to launch at 16. The better reading is that 16 and 13 are tied on the survey, and the in-market test should decide between them.

Running the Analysis With Claude or ChatGPT

This guide hands the analysis to the guide that owns each method, so every prompt lives in one place and gets tested there.

For Gabor-Granger, use Gabor-Granger Analysis with Claude, which handles fixed and branching designs and tests whether the leading price is separable from its neighbor. For conjoint, use Conjoint Analysis with Claude. For Van Westendorp and its extension, use the companion Van Westendorp guide.

Three habits apply to all of them.

Turn on code execution before you paste data, because a model without it estimates instead of calculating. Export one row per respondent with a column for each price answer and each segment you plan to cut. Name the columns plainly, because the prompts in those guides expect them.

Check any interval a model reports against the margin formula in the sample size section. And run any number you’ll report twice, because chat clients don’t expose a seed and two runs can differ.

For the general discipline of checking a model’s analysis, read How to Analyze Survey Data with AI.

Sprig Note: Getting the Data Out. Responses export as CSV or reach Claude, ChatGPT, and other clients through the Sprig MCP connector, with access scoped to the user’s role and response data not used to train models. The connector retrieves at most 1,000 responses per call, so a larger study arrives in pages and needs a completeness check before analysis.

The Critique You Should Know

The standing objection to survey pricing research is hypothetical bias. A respondent who calls a price acceptable has committed nothing, and people state a willingness to pay they wouldn’t act on.

The Evidence on Consumer Goods

Schmidt and Bijmolt (2020) pooled 77 studies and 115 effect sizes, with 24,347 hypothetical and 20,656 real observations, in "Accurately Measuring Willingness to Pay for Consumer Goods: A Meta-Analysis of the Hypothetical Bias," Journal of the Academy of Marketing Science 48, 499 to 518. Hypothetical willingness to pay exceeded real willingness to pay by about 21 percent on average.

The abstract reports three more findings. Indirect methods such as conjoint overstate real willingness to pay significantly more than direct methods do. Bias is greater for higher-valued products and specialty goods. And bias is greater in within-subject designs than in between-subject designs.

The last finding concerns study design. Whether it bears on a Gabor-Granger ladder isn’t tested in the meta-analysis. Voelckner (2006), in Marketing Letters 17(2), 137 to 149, found hypothetical willingness to pay significantly above real willingness to pay for every method tested.

Conjoint Overstated More

Conjoint, widely assumed to be the more conservative instrument, overstated more than direct questions in the pooled evidence. That’s the reason this guide recommends direct methods first and conjoint where the trade-off is the question.

The Counter-Critique

Miller, Hofstetter, Krohmer, and Zhang (2011) compared state-of-the-art approaches against real purchase data in "How Should Consumers’ Willingness to Pay Be Measured? An Empirical Comparison of State-of-the-Art Approaches," Journal of Marketing Research 48(1), 172 to 184.

They concluded: "Even when the OE format and CBC analysis generate hypothetical bias, they may still lead to the right demand curves and right pricing decisions." OE is the open-ended direct question and CBC is choice-based conjoint.

Their incentive-aligned methods, including the Becker-DeGroot-Marschak mechanism, passed both statistical and decision-oriented tests. Bias can be designed down when the respondent has something at stake.

What Incentive Alignment Looks Like

The Becker-DeGroot-Marschak procedure asks the respondent to name a price, then draws a random price. If the draw is at or below the named price, the respondent buys at the drawn price. If it’s above, no sale happens. Because the respondent can’t influence the drawn price, stating a true valuation is the best strategy.

The cost is that a real product and a real payment have to exist inside the study. That’s often impractical for software that hasn’t shipped.

The Limits of the Evidence on Both Sides

The meta-analysis and the head-to-head comparison study different populations and products, so neither settles the question for your category. The pooled finding on direct methods comes from consumer goods, and business software sits outside what it measured. Whether either transfers to business software is untested.

Both point the same way for design. Use the survey to rank, and use a real transaction to confirm.

Where This Leaves the Method

The level is biased and the shape often isn’t. Survey pricing research ranks candidate prices well and forecasts revenue poorly. Use it to choose between prices, never as a discount factor, since 21 percent is an average across studies and not a correction you can apply to yours. And where a live test is possible, confirm the level in market.

Common Mistakes

Nine mistakes recur.

Surveying current customers about a price increase and reading their answers as churn forecasts
Running a Gabor-Granger ladder in ascending order only and reading the falling acceptance as pure price sensitivity
Reporting a perception crossing as the revenue-maximizing price
Writing a vague product description with no billing unit, term, or inclusions
Picking a winner between two prices whose intervals overlap
Mixing prospects and customers in one sample
Treating the survey price as the launch price without a market check
Changing the product description between waves and reading the difference as a market shift
Reading a round-number intersection as precise

Three of them deserve a closer look. Customer surveys on price increases mislead because anchoring and costless outrage distort the answer in both directions. Overlapping intervals mislead because at 100 respondents per price, an eight-point gap in acceptance is frequently noise. And a vague description misleads because respondents price what they imagine, so the range comes out as wide as the category.

The Mistake That Hides Best

The mistake that hides best is letting the screener, the invitation, or an earlier question show a price. A respondent who has seen one number anchors on it, and every later answer inherits it. The model fits and the chart renders, and the curve describes the anchor rather than the respondent.

Read the full respondent path in preview, invitation included. The invitation is the step most often missed, because someone other than the researcher usually writes it.

Pre-Launch Checklist

Run these in order before the study goes live:

Write the decision the study informs in one sentence
Confirm the method matches the branch the decision tree selected
Confirm the product description names inclusions, billing unit, and term
Confirm no price appears in the screener, invitation, or earlier questions
Confirm numeric validation is on for every price-entry question
Preview every branch of a Gabor-Granger ladder
Confirm monadic cells launch together from a randomly split frame
Write the acceptance definition and exclusion rules into the plan
Trace the sample size to the margin tables and the segment plan
Read a small pilot record by record

The last item is the one teams skip, and it’s where a broken pipe or a looping branch gets caught before it costs a full sample.

Synthetic Respondents

A synthetic respondent can produce a plausible set of price answers. Whether those answers carry the price sensitivity of a real population is a different question, and the evidence doesn’t settle it. Bisbee and colleagues (2024) tested language-model "silicon samples" against real survey data in Political Analysis. The synthetic responses varied less than real ones, and small changes in prompt wording shifted the results.

The specific risk for pricing is that a model’s sense of a normal price comes from published prices, reviews, and discussion in its training data. A synthetic panel may hand your category’s list prices back to you. That output looks confident and plausible, which is the direction of error most likely to be believed.

Use synthetic responses to test an instrument before fielding, to catch a confusing question or a broken branch. Don’t use them to produce the range or price a decision rests on. Researchers remain responsible for validating any synthetic output against a real sample.

How Sprig Supports Pricing Research

Sprig supports each stage of this workflow, with limits that are documented rather than inferred. Panels reach prospects, in-product surveys reach customers at a plan page or renewal moment, and email and links reach known lists. Conjoint is a native question type, and the Gabor-Granger and conjoint analysis recipes run in a connected AI client.

The limits belong next to the capabilities. Sprig has no dedicated pricing question type and no documented random assignment of respondents to price cells. Conjoint, MaxDiff, and response quotas are Enterprise features, and conjoint runs on link surveys only. A team that wants a pricing study delivered as a finished dashboard should weigh that.

Panels and in-product surveys also answer different questions. A panel tells you what prospects with no relationship to your product would pay. An in-product survey tells you what the people who use it now would pay, and they’re anchored on what they already see.

Qualtrics documents a Pricing Study solution for Gabor-Granger that ships a survey and a report with a demand curve and a revenue-maximizing curve, set up from a minimum price, a maximum price, and a tolerance. It allows one pricing block per survey. A team that wants the calculation done inside the survey tool, with no analyst in the loop, is better served there today.

Sprig fits a team that can field to the same users it already surveys in product and that analyzes in an AI client. The in-product channel matters most for price changes, because surveying the affected cohort at the plan page and reading answers against behavior in the same product is the workflow the earlier section recommends.

If price in a conjoint is the whole study, specialist conjoint vendors such as Sawtooth Software are built for that depth.

Alternatives and Adjacent Methods

Each of these narrows or confirms a pricing study, and a mature pricing program often uses two or three together.

An in-market price test, where the question is what buyers actually pay. It’s the standard every survey method is judged against
MaxDiff, where the question is which features belong in a tier
Conjoint analysis, where the question is how price trades against features or competitors
A concept test, where the question is whether anyone wants the product at all, since a pricing study assumes the answer is yes
Qualitative interviews, where the question is why a price feels wrong
A fake-door test, which shows a real offer with a price and records how many people try to buy
The Becker-DeGroot-Marschak procedure, where a real product is available and the stakes justify the setup

A fake-door test is closer to behavior than any survey and usually faster to run than a full in-market test.

Frequently Asked Questions

What Is Pricing Research?

Pricing research is survey-based measurement of how potential buyers respond to a product’s price, used to choose a price, range, or structure before committing in market. The main survey methods are Van Westendorp, Gabor-Granger, monadic price tests, and conjoint with a price attribute. Each answers a different pricing question.

How Do You Measure Willingness to Pay?

Willingness to pay is measured four ways. Direct survey questions, such as Van Westendorp, Gabor-Granger, and monadic tests, ask about price outright. Indirect methods, such as conjoint, infer it from choices that include price. Incentive-compatible procedures attach a real purchase to the answer. In-market tests observe what buyers actually pay.

What Is the Difference Between Van Westendorp and Gabor-Granger?

Van Westendorp asks respondents to name four prices, from too cheap to too expensive, and returns an acceptable price range. Gabor-Granger shows specific prices and asks whether respondents would buy, and returns a demand and revenue curve. Van Westendorp finds the range, and Gabor-Granger chooses a price inside it.

How Many Respondents Does a Pricing Study Need?

A pricing study’s sample follows from the precision needed at each price. At 100 respondents per price, an acceptance share near 50 percent carries a margin of error of about ±10 points, and at 400 it’s about ±5. Detecting an eight-point gap between two prices at a 50 percent baseline takes about 606 respondents per price, and each reported segment and each monadic price cell needs that full sample.

Can a Survey Tell You What People Will Actually Pay?

No survey measures what people will actually pay, because every answer is hypothetical. A 2020 meta-analysis of consumer-goods studies found stated willingness to pay ran about 21 percent above real willingness to pay on average. Surveys rank candidate prices well, and an in-market test confirms the level.

Should You Survey Existing Customers About a Price Increase?

Survey existing customers about a price increase only as a directional read. Current customers are anchored on today’s price and report reactions they won’t act on, so behavior after the change is the reliable measure. To test the change itself, use a staged rollout or a holdout group on the affected cohort and read renewal and churn after it ships.

How Do You Test Per-Seat Against Usage-Based Pricing?

Testing per-seat against usage-based pricing is a structure question, and level instruments don’t answer it alone. Describe each structure as a separate product and price each with the same instrument, or run conjoint with structure as an attribute so respondents trade one model against the other directly.

When Should You Use Conjoint Instead of Van Westendorp?

Use conjoint instead of Van Westendorp when price trades against features, tiers, or named competitors in the decision. Van Westendorp prices one product with no alternative in view. Conjoint forces respondents to trade price against what it buys, at the cost of more design effort. Read How to Run a Conjoint Analysis before choosing it.

What Is a Good Optimal Price Point?

No benchmark exists for a good optimal price point, because the result is denominated in your category, currency, and respondents' reference prices. Compare against your own prior studies on the same instrument, and treat the optimal price point as a perception crossing rather than a revenue estimate.

What if You Can’t Reach That Many Buyers?

When the buyer pool is small, size the study to the gap you can actually detect, test fewer prices, or use a range method to find a plausible band and then rely on in-market evidence for the final call. Don’t report a winner between two prices your sample can’t separate.

How Do You Run an In-Market Price Test?

An in-market price test splits real traffic or sign-ups at random between prices and compares conversion and retention. Run it long enough to cover a full buying cycle, keep everything except price identical across groups, and check legal and fairness limits on showing different prices before launch.

Where Does Pricing Analysis Happen?

Pricing analysis for a Sprig study runs in a connected AI client, reached through a CSV export or the MCP connector, rather than inside the platform. Gabor-Granger Analysis with Claude covers that method for Gabor-Granger.