Purchase intent: read the score before choosing the concept
Calculate top-box and top-two purchase intent from fictional counts. Use threshold sensitivity and objections to guide an offer decision without forecasting sales.
Purchase intent measures stated likelihood of choosing a specific offer under stated terms. Read the full answer distribution, price, timeframe and reasons before choosing a concept. “Definitely would” and “probably would” can produce different rankings; use that difference to select the next comparison or pilot rather than convert a survey score directly into sales.
Imagine you are choosing which tea concept to take into a product trial for grocery shoppers. Lumen, the sparkling tea in this illustrative example, gives you a concrete buying situation: a six-pack at $15 in the shopper’s usual store. Concept A leads on positive intent, 50% versus B’s 45%, while B leads on definite intent, 25% versus A’s 18%. Which pattern fits your next investment?
The complete distributions below show the difference. Use them to separate a broad trial proposition from a strong-fit audience hypothesis, then examine the reasons and purchase opportunity that could justify either. The scores supply a choice to investigate, rather than a sales number.
Write an interpretable question
Use the same complete offer and question for each concept.
100 eligible responses per concept
The same answers produce two different rankings.
Intent answer | Results by concept |
|---|---|
Definitely | A (%): 18 B (%): 25 |
Probably | A (%): 32 B (%): 20 |
Uncertain | A (%): 24 B (%): 25 |
Probably not | A (%): 16 B (%): 18 |
Definitely not | A (%): 10 B (%): 12 |
A: 50% positive · 18% definite
B: 45% positive · 25% definite
The five-point distributions show how the ranking changes with the summary you choose. Each contains 100 responses (100%); the percentages are descriptive intent, not a sales forecast.
Imagine this six-pack of Lumen sparkling tea costs $15 at the grocery store you normally use. How likely would you be to buy it within the next four weeks?
Offer “definitely would,” “probably would,” “might or might not,” “probably would not,” “definitely would not” and “not applicable.” Follow with the reason and the product they would buy instead. Someone who rarely shops in that channel should not be forced into the same denominator as an eligible buyer.
A worked intent distribution
For A, the first table contains 18 definitely-would, 32 probably-would, 24 uncertain, 16 probably-would-not and 10 definitely-would-not answers. With 100 eligible responses, the counts are also 18%, 32%, 24%, 16% and 10%.
The top-two-box share is (18 + 32) / 100 = 50%. The top-box share is 18 / 100 = 18%. Both calculations are correct, but they emphasize different strength of intention. Keep the full distribution so the reader can see that difference.
See why the two scorecards disagree
Concept B has 25 definitely, 20 probably, 25 uncertain, 18 probably not and 12 definitely not responses. Its top-two score is 45%, lower than the first concept's 50%, while its top-box score is 25%, higher than 18%.
The descriptive ordering changes with the threshold: B has fewer positive answers and more definitely-would answers in the example. That alone does not establish a stable audience tradeoff. If these were independent random human samples of 100 each, the approximate 95% interval for A-minus-B top-two intent would be −8.8 to +18.8 points; for B-minus-A top-box intent, −4.3 to +18.3 points. Both include zero, before multiple-comparison adjustment. Investigate reasons and segments as hypotheses. These human-sampling calculations do not describe synthetic-model uncertainty. See NIST’s two-proportion guidance.
Carry intent into a realistic purchase opportunity
People may like an idea but never encounter the offer, switch away when a competitor discounts, or discover that the experience does not meet expectations. The study also makes attention and consideration easier than many actual shopping situations.
Calibration requires appropriate observed evidence. If you have historical concept studies and subsequent market outcomes, compare them carefully within relevant categories, instruments and launch conditions. A universal multiplication factor such as “intent × 0.5” is not justified merely because it makes the estimate look conservative.
Synthetic intent adds a further distinction: the response is modeled. Neither the count of generated responses nor a polished probability label converts it into independent human purchasing evidence.
Separate rejection, inapplicability and a missing answer
The distributions above contain 100 valid eligible answers per concept. Real instruments often have other response states. Keep those outside the five-point distribution until you define the base.
Response state | Meaning and reporting |
|---|---|
Definitely would not | Meaning: Eligible person rejects the shown offer Reporting treatment: Retain in the ordered intent scale |
Not applicable | Meaning: Scenario does not fit the person Reporting treatment: Report separately and review eligibility |
Cannot answer because price is absent | Meaning: Offer information is insufficient Reporting treatment: Diagnose the stimulus, rather than treat as rejection |
Blank or technical failure | Meaning: No valid intent evidence Reporting treatment: Report missingness and exposure status |
Do not improve the positive share by dropping valid negative answers. Equally, forcing a person outside the grocery scenario to express an intent creates misleading precision. Publish the eligible assigned base, valid scale base and excluded states so the summary can be reconstructed.
Use reasons to decide which uncertainty matters next
For a diagnostic follow-up, “probably would, if my store stocks it” identifies availability; “definitely would not because I dislike sparkling drinks” identifies category rejection; and “might or might not until I know sweetness” identifies missing offer information. These meanings call for different actions even though all can affect the positive share.
Keep the reason, named alternative and occasion linked to each intent answer. Adding a missing sweetness description can clarify the existing tea; changing its recipe creates another offer. The analyst should show which proposed change addresses an actual condition behind the response.
A disagreement between scores is a useful question
A contains 32 “probably would” answers; B contains 20. B contains 25 “definitely would” answers; A contains 18. The next analysis should explain those two concentrations. Are A’s conditional buyers waiting for a stocked outlet, or merely open to many drinks? Are B’s definite buyers attached to a routine the launch can serve? The displayed counts contain no linked reasons, so both explanations remain hypotheses.
For a small prototype intended for one recurring occasion, investigate B’s definite buyers first. Progress B only if their reasons identify that occasion, the audience is reachable and the offer can deliver what they expect. For a launch seeking wider paid trial, investigate A’s additional conditional buyers and the conditions needed to make buying possible. A larger positive share matters commercially only when those conditions can be supplied.
Agree that objective before seeing which cut favors the team’s preferred concept. If neither pattern can be connected to a feasible audience and use occasion, retain both candidates or revise the offer. Switching between top-box and top-two after the readout would change the question rather than resolve it.
Put stated interest beside the actual opportunity
The next pilot should preserve the offer people evaluated: the same quantity, price and information, available in a defined channel. Record who had an opportunity to purchase and the outcome over the stated window. That makes differences between consideration, checkout and paid trial interpretable.
Choose a pilot comparison that can overturn the provisional choice. For the recurring-occasion route, record paid trial and whether use returns in that routine. For the broad-trial route, record the actual opportunity to buy and paid trial under the complete terms. Suitable historical calibration can inform planning; without it, keep the survey distribution and observed outcomes separate.
Use intent to make the next decision
A’s breadth and B’s strength deserve different briefs.
Ask the analyst for one audience-linked comparison: the full intent categories, reasons, alternatives and purchase conditions for the first audience the business could serve. Lead the recommendation with the candidate that fits the declared commitment and the fact that would reverse it. The illustrative intervals above leave the small overall gaps unresolved; that uncertainty belongs beside the choice.
A completed recommendation can therefore be modest and useful: develop a candidate for a defined prototype, revise a condition that blocks consideration, or take both viable offers into a focused pilot. “A wins 50% to 45%” omits the very tradeoff the team needs to decide.
Sapien can compare modeled intent with comprehension, audience differences and objections, then recommend how the offer should progress. Use our concept-testing guide to prepare consistent questions or discuss a study with Sapien. If the unresolved variable is price, begin with our pricing research guide.
