Gabor–Granger analysis: demand curve and revenue-maximising price

André Gabor and Clive Granger asked shoppers in 1961 whether they would buy a product at a given price, then repeated the question at other prices. The result is a buy-response curve: the share of people who accept each price. Multiply it by the price and you have a revenue curve, which is the closest thing a survey gets to the demand curve an economist means.

Two properties make this useful. Revenue has a peak: at low prices you keep almost everyone but earn little per sale, at high prices the reverse, and somewhere between them the product of the two is largest. And elasticity crosses −1 exactly at that peak — above the peak a 1% price rise loses more than 1% of buyers, below it less. If your ladder shows elasticity nowhere near −1, the peak is outside the prices you tested.

The margin curve moves the answer up

The revenue peak ignores cost, so it is the right answer only if serving a customer is free. Add a variable cost and the optimum moves to a higher price, because the buyers you shed at the bottom of the curve were the least profitable ones. In the example on this page revenue peaks a rung lower than margin does — a gap worth checking before you quote the “Gabor–Granger optimal price” to anyone.

Gabor–Granger and Van Westendorp answer different questions

Van Westendorp’s price sensitivity meter asks where a price stops making sense and gives you a range plus how much your market disagrees. Gabor–Granger asks whether people would buy and gives you a curve with a revenue peak. Van Westendorp is better for an unfamiliar product where you do not yet know the order of magnitude; Gabor–Granger is better once you have a range and need to choose inside it. Running Van Westendorp first to set the ladder, then Gabor–Granger inside it, is the standard pairing.

This method produces a confident-looking curve out of stated intentions, which is exactly the combination that gets teams into trouble.

• It systematically overstates willingness to pay. Saying yes in a survey costs nothing. There is no budget, no approval process, no competitor offering the same thing cheaper, and no moment where the card comes out. Treat the shape of the curve as informative and the level as optimistic.
• The peak can only ever be a price you tested. The answer is a rung, not a number: a ladder of 19/29/39 cannot tell you that 34 is better. And if the peak lands at the first or last rung, the real optimum is outside your range and the tool says so — extend the ladder rather than reporting the edge.
• The ladder direction biases the answers. Ascending ladders produce higher acceptance (people anchor on the first low price and keep agreeing); descending ladders produce lower. Randomising the start is the fix, and if your survey did not do it, the level of the whole curve is suspect.
• It assumes no competitors and no alternatives. The respondent is pricing your product in a vacuum. A curve collected before a competitor’s price cut describes a world that no longer exists.
• Acceptance is not volume. “Would buy” multiplied by an assumed number of buyers is not a forecast. The buyers number you type is an assumption; the revenue curve inherits all of its error, and its shape — where the peak is — is the only part that survives a bad estimate.
• Rising acceptance means the data is too thin. Demand cannot genuinely increase with price in this design. When it appears, it is sampling noise from too few respondents per rung, and the monotone repair on this page shows you the nearest coherent curve rather than pretending the wiggle is a finding.
• One price for everyone hides the packaging question. A single ladder averages over segments that may want very different prices. If your buyers split into groups, run the ladder per segment before concluding anything about one price.
• It says nothing about churn. The price that maximises revenue on first purchase may maximise cancellation three months later, and no survey question here can see that. For subscriptions the decisive evidence is retention at the new price, not acceptance in a questionnaire.
• Elasticity from six rungs is not an elasticity. The arc elasticity between two prices in a survey is a descriptive number about your sample, not a market parameter you can plug into a model.

Used honestly, Gabor–Granger narrows a pricing argument from “what should we charge?” to “which of these four prices, and what do we watch after we pick?” That is a real contribution. It is not a price recommendation.

• Bracket it with Van Westendorp first. The price sensitivity meter gives the range that the ladder should cover, and shows whether your market is really one market.
• Check the price survives the unit economics. Run the winner through the CAC payback period calculator and the LTV:CAC calculator: a price that maximises revenue per buyer can still fail to pay back acquisition.
• Test it on real traffic, not more respondents. A pricing-page experiment beats any survey. The A/B test calculator sizes it; the sample ratio mismatch calculator tells you whether you can trust the result.
• Watch what happens after the sale. A higher price changes who signs up. The retention cohort calculator and the trial-to-paid conversion calculator show whether the new cohort behaves like the old one.
• Decide what belongs in the plan. Willingness to pay depends on the package: the Kano model survey analyzer separates features people expect from features they would pay extra for.

1. Gabor A., Granger C. W. J. “On the Price Consciousness of Consumers.” Applied Statistics 10(3), 1961 — the original buy-response ladder.
2. Gabor A., Granger C. W. J. “Price Sensitivity of the Consumer.” Journal of Advertising Research 4, 1964 — reading the ladder as a demand curve.
3. Lipovetsky S., Magnan S., Zanetti-Polzi A. “Pricing models in marketing research.” Intelligent Information Management 3(5), 2011 — comparison of Gabor–Granger and the price sensitivity meter, including the tendency of stated-price methods to overstate acceptance.
4. Barlow R. E., Bartholomew D. J., Bremner J. M., Brunk H. D. Statistical Inference under Order Restrictions. Wiley, 1972 — pool-adjacent-violators, used for the monotone repair.