How to Test Pricing Before Launch: Conjoint, Willingness to Pay and Annual or Monthly Plans
On this page
Test pricing as a choice, not as a number. Put complete offers in front of buyers, vary the price together with what the buyer gets, and measure which offer they pick. A choice experiment of that kind answers the question a launch turns on: at this price, with this package, against these alternatives, does the buyer choose us? Van Westendorp answers a different question, what price range feels acceptable for a fixed offer; Gabor-Granger answers another, how stated purchase likelihood falls as the price of a fixed offer rises. Use those when the offer is not changing.
Subconscious can run that choice experiment on a simulated market before you commission human research. It randomizes price and package across simulated buyers, estimates the effect of each price step on modeled choice with a confidence interval (method described on fidelity.subconscious.ai), and reports which option leads. The public evidence says what that result is good for. The July 2026 working paper (not peer reviewed, summarized on fidelity.subconscious.ai) reports a mean Spearman rank correlation of estimated choice parameters of 0.73 across 43 design-filtered replications of published choice studies and 0.55 across roughly 300. In the public study registry on 2026-11-12, rank was graded "holds" in 103 of 308 detailed comparisons and direction in 127. In the 27 studies that state a size ratio, the simulated effects ran from 1 percent of the human size to 3.8 times it, and the registry graded none of those sizes equivalent to the human result. So treat the simulated ranking as the set of candidates to test, and get direction, order and size for the leading offers from a human sample before any revenue forecast. The validation page holds the study-by-study record, misses included.
Which pricing method answers which question
| Question you need answered | Method | What it measures | What it cannot tell you |
|---|---|---|---|
| What price range feels acceptable for this product as it is? | Van Westendorp Price Sensitivity Meter | Four perceived thresholds (too cheap, bargain, expensive, too expensive) and where their curves cross | Whether anyone would pay, or how a feature change moves the price |
| How does stated purchase likelihood fall as the price rises, for a fixed offer? | Gabor-Granger | A stated demand curve at specified prices, and a revenue-maximizing price for that offer | Trade-offs against competing offers or against features |
| Which package, at which price, do buyers choose against the alternatives? | Choice-based conjoint (a discrete choice experiment) | Estimated effect of each price and feature level on choice; preference share for whole offers | Realized sales |
| Which price structure should we take into a human study or a live test? | Simulated choice experiment (Subconscious) | The same kind of estimates as conjoint, on simulated buyers, before anyone is recruited | Which candidates survive a human test |
| What happens to revenue when we change the live price? | Live price test with a holdout | Observed purchases and revenue | Why buyers moved; hard to run before launch |
Conjointly, a survey platform, positions its Van Westendorp product as finding the price range buyers find acceptable for one product, and its Gabor-Granger product as estimating elasticity and the revenue-maximizing price for one product (products page, no date shown, read 2026-11-12). Sawtooth Software notes that choice-based conjoint "is known as discrete choice modeling (DCM) or discrete choice experiments (DCE)" (CBC page, no date shown, read 2026-11-12). Both vendors sell conjoint analysis for the trade-off question. The Sawtooth comparison and the Conjointly comparison set out how a simulated choice experiment differs from a fielded one on cost, time and evidence.
Worked example: annual and monthly billing at Finta
Finta, a fundraising platform for private companies, charged $348 per year with no monthly option. After modeled pricing work with Subconscious, its CEO reported in an email dated 2024-10-01 that the company moved to a $950 annual plan and reintroduced a $99 monthly plan. Five customers took the monthly plan. The CEO also reported closing customers at the old $348 price, which read as a discount with $950 as the anchor on the sales call (Finta case, source email dated 2024-10-01).
The case page records the modeled result (simulated demand supported a two-tier structure at $99 per month or $950 per year), the adopted prices and the monthly-plan uptake. The adoption is a client-reported pricing decision, not a controlled estimate of revenue lift: the annual price rose from $348 to $950, 2.73 times the original, calculated from the reported prices; the revenue effect is not documented. As an illustration of how to frame such a test, make the two menus the experimental conditions: annual-only in one arm and annual-plus-monthly in the other, with fully priced plans and a no-purchase option in each, and measure which menu more buyers choose from, not "$348 or $950".
Worked example: where the simulation missed, at Mintel
Mintel ran a facial-moisturizer conjoint with two waves of 1,000 human respondents and had Subconscious run the same attribute set on simulated buyers built from demographics alone, with no human preference or choice data as input. Modeled attribute distributions reached 0.79 to 0.88 Jensen-Shannon similarity against the human results; that is distribution agreement, not a percentage of correct predictions (Mintel case).
The price attribute is where the two diverged. Human demand dropped sharply between $34 and $48; the model assumed a smoother decline. For a pricing test that is the limit to design around: the simulation missed a cliff in the price curve that only the human wave showed. Test the price grid densely near the candidate price points. The Mintel case records the comparison.
Worked example: a price point for an existing product
New Age Floral, selling an existing product with flat sales, used five simulated experiments to test price alongside compostability wording. The simulator favored $60 for maintaining modeled share and estimated declining demand above $100. "Home compostable" produced higher modeled purchase likelihood than "commercially compostable". These are modeled findings, not observed sales thresholds, and the case does not document a post-study sales result (New Age Floral case, source document dated 2026-05-15).
How to set up the test
- Write the decision as a choice. Name the offers you could ship, including the one you ship if you do nothing. "Raise the monthly price" and "add an annual tier" are different interventions with different comparisons.
- Define the buyer population and the outside option, so the study does not overstate demand.
- Fix the price grid before the run and keep it dense near the prices you would set. Mintel's human and modeled price curves diverged between $34 and $48.
- Randomize price with the package, so each effect is estimated from designed contrasts rather than from a one-off price change.
- Set the smallest useful difference in advance: the gap in modeled choice, at the cost of implementation, that would change the price you ship.
- Read the ranking and the direction from the simulated study as candidates, then run a human choice study on the leading offers, sized from the smallest useful difference in step 5. The human study checks direction and order as well as size. A revenue forecast then needs the market, availability and repeat-purchase inputs the study compatibility guide lists, and its own validation.
Willingness to pay
A choice experiment yields a willingness-to-pay figure only when price enters the model as a currency amount with a valid signed slope. Subconscious's study compatibility guide (updated 2026-10-23) says its legacy willingness-to-pay tool returns categorical utility ratios, not currency willingness to pay, and that a currency figure requires a valid signed price slope in currency units. Ask for the currency price model and its uncertainty when you need a dollar figure, and remember that a modeled willingness to pay is not an observed payment.
To scope a pricing study, book a decision review and bring the offers, the buyer definition and the decision deadline.
How this page was made
By: Avi Yashchin, CEO and Co-Founder. Drafted with AI assistance on 2026-11-07 and updated on 2026-11-12. Customer figures come from the public case pages linked beside them; validation figures come from the fidelity.subconscious.ai summary of the July 2026 working paper and the public study registry, read on 2026-11-12. Competitor method descriptions come from the vendors' own product pages, read on 2026-11-12.