How Can You Test Whether a New Product Will Sell — Before You Launch?
Conjoint and concept tests ask what people say they'd choose. What predicts a launch is what actually drives choice — often something customers can't articulate.
The short answer
Most pre-launch research asks people to rate hypothetical options — that's what conjoint and stated-preference concept testing do. The trouble is that what customers say they'd choose diverges from what they actually do. That gap between stated preference and real behaviour — the say-do gap — is exactly why so many products that "tested well" still flop. To predict whether a product will sell, you have to measure the true drivers of choice, which are often non-conscious, model them causally, and simulate the launch.
Why stated-preference testing misleads
Conjoint and concept tests put respondents in front of hypothetical trade-offs and ask them to rate or rank. The answers feel like data, but they capture only what people can consciously justify. Ask someone why they'd buy a new financial product and they'll tell you it's "secure" and "simple" — because those are the reasons that are easy to say out loud. The factors that genuinely move their choice tend to be emotional and non-conscious, and a conjoint on stated attributes never sees them. So a concept can score beautifully on the things people say matter, then fail once it reaches a real market driven by things they never mentioned.
This is the same failure we describe in why market research is broken: measuring what's visible and stated, not what actually causes behaviour.
What customers said mattered wasn't what drove interest
In a case from Dr. Buckler's book The End of the KPI-Illusion, a mobile-payment / digital-wallet was tested before launch. Asked directly, customers said security and simplicity mattered most. But a causal analysis using Universal Structure Modeling told a different story.
Because relative advantage was itself caused by emotional uniqueness, the winning go-to-market was to market the product emotionally — not on security features. A stated-preference test would have doubled down on exactly the wrong message.
Stated preference vs the real drivers of choice
| Dimension | Conjoint / concept test | Causal driver simulation (SUPRA) |
|---|---|---|
| What it measures | What people say they'd choose | What actually drives choice — including non-conscious factors |
| Method | Rating hypothetical trade-offs | Causal AI (USM) on the true drivers of interest |
| Risk | The say-do gap — tests well, flops in market | Captures emotional & indirect drivers customers can't state |
| Output | Preference scores for stated features | A simulated launch with the real levers of demand |
"When you ask people what they'd choose, they answer with the reasons they can defend — not the ones that actually move them. Stated preference is a story customers tell about themselves; real behaviour runs on something they can't put into words."
How to actually predict a launch
The pattern generalises well beyond one digital wallet. To predict whether a product will sell, measure the true drivers of choice and simulate the launch, rather than trusting stated preference. In practice that means:
- Measure the non-conscious drivers — capture what customers respond to, not just what they can name, using implicit measurement.
- Model the drivers causally — separate the factors that cause interest (like emotional uniqueness) from the ones customers merely cite, and trace how they feed each other.
- Simulate the launch — estimate how demand responds to each lever, so you know which message and which feature actually move the market before you spend on it.
The method is the same causal AI foundation behind our innovation and product research work, and the reasoning is developed in Dr. Buckler's book The End of the KPI-Illusion.
Frequently asked questions
How can I test whether a new product will sell before we launch it?
Don't rely on asking people to rate hypothetical options. Stated preference — what customers say in a conjoint or concept test — diverges from what they actually do, so a product can test well and still flop. Instead, measure the subconscious drivers of choice, model them causally to find which truly move interest, and simulate the launch. In the digital-wallet case from The End of the KPI-Illusion, customers named "security and simplicity," but causal analysis showed lifestyle compatibility drove roughly 50% of product interest.
Is there research more reliable than conjoint for predicting new-product success?
Yes. Conjoint and stated-preference concept testing capture what customers can articulate — not the non-conscious factors that drive real choice. Causal modelling of the true drivers is more reliable because it captures what people can't or won't say. Using Universal Structure Modeling, the digital-wallet analysis found perceived relative advantage was itself caused by emotional uniqueness, and lifestyle compatibility drove about half of interest — insights a conjoint on stated features would never surface.
Why do products that test well still flop?
Because the test measured stated preference, not real behaviour — the say-do gap. When you ask people to rate options, they report the reasons they can consciously justify (like security or simplicity), while the factors that actually drive their choice are often emotional and non-conscious. A concept that scores well on stated attributes can miss the real levers of demand entirely, and fail in market.
What actually predicts new-product success?
The true, often non-conscious drivers of choice — measured, modelled causally and simulated as a launch — rather than the reasons customers state. In the digital-wallet case, the winning move was to market the product emotionally rather than on security, because emotional uniqueness drove perceived relative advantage and lifestyle compatibility drove roughly 50% of interest. Predicting success means measuring what really moves choice, not what people say moves it.
Find out what your concept test isn't telling you
Bring your new-product concept. We'll show you whether a causal read reveals the real drivers of choice your stated-preference test is missing.
Get my AI Diagnostic →