Guide
Concept testing, frameworks, methods & how to validate before building
A practical guide to concept testing, for products, creative, messaging, and pricing. Methods, sample size, and the pitfalls of asking 'would you buy this?'
Concept testing puts an unfinished idea in front of its target audience to see whether the idea itself holds up before money goes into building or shipping it. It is not the same as creative testing, which evaluates execution (the ad, the headline, the page). Concept is the idea. Creative is the craft. Mix them up and you will kill a strong idea because the mockup looked rough, or ship a bad idea wrapped in a beautiful design.
A concept test should be designed to find what is wrong with the idea, not to collect compliments. If the test cannot fail, it is theatre.
What you can concept-test#
Most teams default to product features. The aperture is much wider than that.
Product
A new product, feature, or workflow. Test the underlying job before the spec is written.
Message
Value propositions, taglines, category framing. Find the wording that lands with strangers, not the one your team voted for.
Creative
Ad angles, hero ideas, campaign concepts. Test the angle before producing the assets.
Brand
Names, identity directions, repositioning. Check that the new direction reads the way you intend it to read.
Feature priority
A backlog of ideas ranked by what your audience actually wants, using methods like MaxDiff.
If you are testing a new pricing page, show two prospects each version and ask what they think they are paying for. If they cannot tell you, the page has not done its job, regardless of how the visual scores.
The "would you buy this?" trap#
The most common concept-test question is also the worst predictor. Asking "would you buy this?" or "how likely are you to use this?" produces inflated answers because there is no money on the table, no competing alternative, and a polite social pressure to be encouraging.
Decades of new-product research show that stated purchase intent overestimates real demand. The exact discount factor varies by category, question wording, and channel, but the direction is consistent: top-box intent ("definitely would buy") is a soft, optimistic number that has to be heavily discounted before it predicts behavior. Treating a high intent score as a green light is how teams ship products that nobody buys.
Two practical implications:
- Use intent only to compare concepts against each other, not as an absolute forecast.
- Pair every intent question with a behavioral or trade-off question that costs the respondent something to answer.
Better questions to ask#
Behavior beats opinion. Trade-offs beat ratings. The questions below are slower to analyze but a lot more honest.
- What do you do today to solve this?:
Past behavior is the strongest signal of future behavior. If the answer is "nothing," the problem is probably not as painful as the concept assumes.
- What would this replace?:
Forces a real-world comparison and surfaces the actual competitive set. The competition is often a spreadsheet, a junior teammate, or doing nothing.
- Walk me through the last time you needed this:
Specificity is the tell. Detailed stories with names, dates, and tools signal a real use case. Vague answers signal a polite respondent.
- What would have to be true for you to pay for this?:
Surfaces the real conditions: integrations, proof, approvals, guarantees. More useful than a willingness-to-pay scale.
- What did you think this was, before I explained it?:
Catches positioning failures fast. If the first read is wrong, the framing is wrong, even if the underlying concept is fine.
For more on the behavioral vs attitudinal split, see qualitative vs quantitative research and the discovery patterns in customer discovery.
Main methods#
There are four workhorse methods. Pick by what decision the test has to support.
Monadic
Each respondent sees one concept and answers questions about it. Cleanest read on a single concept's standalone appeal because there is no contamination from comparison. Higher sample cost, since you need a full cell per concept.
Sequential monadic
Each respondent sees several concepts in sequence and evaluates each on its own. Cheaper than pure monadic and the right default when you are shortlisting two to four candidates. Randomize order to control for position effects.
MaxDiff (best-worst scaling)
Show small sets of features, messages, or benefits and ask which is most and least appealing. Produces a ranked, comparable list that is much sharper than rating each item on a 1 to 5 scale where everything ends up a 4. Use it when you need to prioritize a long list.
Van Westendorp for pricing concepts
Four price-sensitivity questions ("at what price is it too cheap to be credible, a bargain, starting to feel expensive, too expensive to consider"). Use it on pricing concepts, not for setting list price to the dollar. Pair with pricing research when the decision is the actual number.
A useful pattern: qualitative interviews to figure out the right concepts, sequential monadic or MaxDiff to compare them at scale, then a focused intent or behavioral measure on the winner.
Sample size: depth vs validation#
Sample depends on what the test has to prove. Three honest defaults:
- Qualitative depth, around 10 to 30 conversations per segment. Enough to reach thematic saturation, where new interviews stop producing new insight. This is the work that explains why a concept lands or fails, and it is what you need before any quant test is worth running. The mechanics overlap heavily with user research.
- Quantitative validation, roughly 150 to 300+ per cell. Enough to compare concepts statistically, detect meaningful differences in preference, and segment results. MaxDiff and pricing studies sit at the higher end.
- Internal directional reads, 30 to 80 responses. Useful for a gut check across a known audience when you are not planning to publish numbers. Be honest about confidence.
The mistake is running a 400-respondent quant test on the wrong concepts. Precise measurements of bad options do not make better decisions. Do the qualitative work first, then validate.
If you want to run a quick qualitative reaction test on a concept, AI-moderated interview tools like Diaform can collect open-ended reactions and synthesize them across respondents, which compresses the slowest part of the loop without losing the depth.
One closing thought#
A concept test is not a verdict. It is a structured way to learn what would have to change for the idea to work. Treat the output as a list of risks to retire and conditions to meet, not a thumbs up or thumbs down, and the test will earn its place in the cycle every time.