Gabor-Granger's Fatal Flaw: It Anchors the Answer It's Trying to Measure
Show a price, ask "would you buy?", raise or lower, repeat. You get a clean demand curve — built on a sequence of answers the method itself contaminated.
What the Gabor-Granger method actually does
Gabor-Granger is one of the oldest and most-used pricing techniques in the survey toolkit, and for good reason: it is simple, fast, and produces a number a manager can act on. Here is how it works, fairly.
You show a respondent a single product, described as clearly as you can, at one specific price. Then you ask the only question that matters: at this price, would you buy it? Usually the answer is a purchase-likelihood rating or a yes/no. Then you change the price — up or down — and ask again. Across the sample, you vary the price across a handful of points. When you aggregate the "yes" rates at each price, you get the share of people who would buy at every price level. That share, plotted against price, is a demand curve.
And because you now have demand at each price, you can multiply price × demand to get expected revenue at each point. The price that maximizes that product is your revenue-maximizing price. That single output — "charge $X" — is why the method has survived for sixty years.
Worked mini-example: a subscription
You're pricing a SaaS subscription and test four points. Each respondent sees one price and says whether they'd subscribe. The aggregated "would buy" rates come back like this:
Demand falls as price rises — the expected, downward-sloping curve. But revenue peaks at $12, not at the cheapest or the most expensive point. Gabor-Granger hands you "$12" and a tidy chart to defend it. The question is whether that number means anything.
Where Gabor-Granger is genuinely useful
Credit where it's due. For a single, well-understood product in an existing category — where buyers already have a reference price in their heads — Gabor-Granger gives you a fast, cheap read on elasticity. It is far simpler to field and analyze than conjoint, it needs no specialist software, and respondents understand the task instantly. If you need a directional sense of how sensitive demand is to price, and you need it this week, it does the job. It is a smell-test, and a decent one.
The trouble starts the moment anyone treats the output as a forecast.
The turn: it anchors the very thing it measures
Here is the structural problem. Gabor-Granger measures willingness to pay by showing people a price. But the moment you show a price, you have anchored the respondent. The number you put on screen becomes the reference point against which they judge "fair." You are not reading their willingness to pay — you are partly creating it.
It gets worse when you walk a single respondent up or down a price ladder. The order in which prices appear biases the curve: ascending sequences and descending sequences produce different answers from the same people. And underneath both problems sits the oldest one in survey research — answering "would you buy at $15?" on a questionnaire costs the respondent nothing. There is no budget, no checkout, no regret. This is the say-do gap, and it hits pricing harder than any other research domain, almost always biasing stated intent away from real behavior.
So Gabor-Granger measures stated, anchored, consequence-free intent. It does not measure revealed willingness to pay. Those are not the same animal, and the gap between them is where margins go to die.
Three biases, concretely
1. Price anchoring. The price you display sets the respondent's reference point. Show $19 first and $12 feels like a bargain; show $9 first and $12 feels expensive. The "demand curve" partly reflects the prices you chose to test, not an independent truth about the market.
2. Sequence / order effects. Walking respondents up an ascending ladder yields different acceptance than walking them down a descending one. People resist the first "no" and then stay consistent, so the starting direction drags the whole curve. Randomizing helps, but it spreads the bias across the sample rather than removing it.
3. Hypothetical bias — no real budget. A survey "yes" is free. No money leaves the account, no trade-off against rent or a competitor's offer is forced. Stated purchase intent routinely overstates real conversion, and by an amount that is itself unstable across categories. You cannot simply subtract a fixed correction.
What to do instead
If the decision matters, anchor your estimate to behavior, not to a survey screen.
- Derive elasticity from real data. If you have transaction history, past price changes, or behavioral data, the demand response is already written in what people actually did. Read it there first.
- Use implicit measurement for gut-level price-value fit. Reaction-time-based methods capture the unconscious "is this worth it?" association before deliberate, anchored reasoning kicks in — closer to how buyers actually decide at the shelf or the checkout.
- Use Causal AI to estimate the counterfactual. The real question is never "what would people say at $X?" It is "what would demand do if we moved price from A to B?" That is a causal question, and stated-preference methods can't answer it. SUPRA models the causal drivers of price acceptance on your existing market and behavioral data, so you're forecasting a move, not interpreting a guess.
And to be honest about it: when the budget is genuinely tiny and the stakes are low, Gabor-Granger is acceptable as a directional smell-test. Field it, read the curve, and treat the result as a hypothesis to validate — not as the answer. Just don't bet a pricing strategy on it.
"Every pricing method that shows people a number is, in part, writing the answer it claims to read. The discipline isn't in the curve. It's in refusing to trust it until behavior confirms it."
When is Gabor-Granger OK — and when isn't it?
| Your Situation | What to Use |
|---|---|
| Single product, existing category, tiny budget, directional answer only | Gabor-Granger OK (as a smell-test) |
| You have transaction or past-price-change data | Derive elasticity from behavioral data |
| Premium positioning where stated intent underestimates | Implicit measurement, not Gabor-Granger |
| You need to forecast a specific price move | Causal AI — counterfactual, not stated |
| The cost of being wrong is high | Behavioral + Causal combined |
| Multi-feature product, price is one variable among many | Conjoint as a start, validated causally — not Gabor-Granger |
Frequently asked questions
What is the Gabor-Granger method?
A survey-based pricing technique. Respondents are shown a product at a specific price and asked whether they'd buy it. The price is raised or lowered across a sequence of points, and aggregating the "yes" rates produces a demand curve and a revenue-maximizing price. It works best for a single, well-understood product in an existing category.
Gabor-Granger vs. Van Westendorp — what's the difference?
Gabor-Granger asks a binary purchase question at specific price points and outputs a demand and revenue curve. Van Westendorp asks four perception questions ("too cheap", "cheap", "expensive", "too expensive") and outputs an acceptable price range, not a demand curve. Gabor-Granger is better for optimizing a single price; Van Westendorp for bounding a range. Both are stated-preference methods and share the say-do gap.
Is Gabor-Granger accurate?
Directionally useful but systematically biased. Showing a price anchors the respondent, the ascending or descending sequence biases the curve, and a hypothetical "would you buy?" has no consequence — so it measures stated, anchored intent rather than revealed willingness to pay. Treat it as a smell-test, not a forecast, and validate against real transaction or behavioral data before betting margin on it.
How many price points should you test?
Typically five to seven price points spanning a realistic range around your expected price, with enough respondents per point for stable estimates. Too few and the curve is coarse; too many and respondent fatigue and sequence effects grow. Randomizing the starting point and order across respondents reduces — but does not remove — anchoring and sequence bias.
Get a pricing-research diagnostic
If Gabor-Granger handed you a number and your gut says don't trust it, your gut is right. We pressure-test pricing research against behavior — before you commit margin to a survey artifact.
Get my AI Diagnostic →