Confirmation Bias in Product Discovery: How Founders Interview Customers Into Lying

Confirmation bias in product discovery does not announce itself. It arrives as a warm conversation, a nodding customer, and a founder walking back to the office certain the market has spoken. The market has not spoken. The founder has. The customer, sensing enthusiasm and social obligation, has simply agreed. The result is a discovery process that feels rigorous and produces nothing reliable.

The fix is not more interviews. It is a structural change in how founders frame questions, select interviewees, and interpret what they hear — before a single line of code is written or a dollar of runway is spent.

Why the most motivated discoverers are the most biased

In 1960, British psychologist Peter Wason designed a deceptively simple experiment. ¹ He showed participants the sequence 2, 4, 6 and asked them to discover the underlying rule by proposing their own sequences. ¹ People tested triples that confirmed their guessed rule and rarely tested sequences that could break it. Only a disconfirming test — proposing something that should fail — could reveal the true rule, which was simply “any ascending numbers.” ¹ The experiment became one of the most replicated findings in cognitive psychology and the foundation of confirmation bias research. ²

The mechanism Wason identified operates at three compounding stages: biased search (seeking confirming evidence), biased interpretation (reading ambiguous evidence as supportive), and biased memory (selectively recalling confirmations). ³ These stages do not operate sequentially. They run in parallel, and they are largely independent of cognitive ability — meaning that a brilliant, well-intentioned founder is just as susceptible as anyone else. ³

Now place that founder in a customer interview. She has spent months on an idea she believes in. She has a script. She has hypotheses. And she has, almost certainly, already decided what the customer is going to tell her. ⁴ With the right framing, everything can be interpreted as a green flag. ⁴ The interview does not surface truth; it surfaces the founder’s prior beliefs, laundered through a stranger’s politeness.

The social contract that makes customers lie to protect you

Confirmation bias on the founder’s side is only half the problem. The other half sits with the customer. Social desirability bias is a systematic research error in which participants present answers that are more socially acceptable than their true opinions or behaviours. ⁵ In qualitative research — which is precisely what a customer interview is — this bias can lead to distorted conclusions about the phenomenon being studied. ⁵

The mechanism is not deception. It is social instinct. Participants tend to overestimate socially acceptable behaviours and underestimate true opinions when those opinions are socially undesirable. ⁶ In a face-to-face interview, these pressures are especially salient: research published in Political Science Research and Methods found that respondents significantly underreport negative emotions when interviewed face-to-face compared to anonymous modes. ⁷ Translate that to a founder interview: the customer underreports scepticism, underreports indifference, and overreports enthusiasm — not because they are dishonest, but because the social context demands it.

Rob Fitzpatrick, in The Mom Test, diagnosed this dynamic with precision: if you ask your mother whether your business idea is good, she will say yes — she loves you and does not want to hurt your feelings — and your customers, without even realising it, do the same thing. ⁸ The book identifies three categories of bad data that result: compliments, hypothetical fluff, and wishlists — each of which leads founders to chase false signals, build the wrong product, and waste time and money. ⁹ Incorrect information and false positive feedback, Fitzpatrick argues, are more dangerous than direct bad news. ¹⁰

How the question structure does the lying for you

Most interview bias is not introduced by the founder’s attitude. It is introduced by the founder’s grammar. ¹¹ Even well-trained researchers fall into what behavioural scientists call “question bias” — the tendency for questions to prime specific types of answers. ¹² A scripted interview guide that asks “How satisfied are you with our onboarding?” will produce different insights than one that asks “Tell me about the last time you set up a new tool at work.” ¹²

The structural errors are consistent across industries and geographies. ¹³ Founders ask leading questions that tell the interviewee the answer they want. They talk about their solution too early, converting a discovery conversation into a sales pitch. They accept surface answers without probing — and the first answer is often what people think you want to hear. ¹³ They ask hypothetical future questions (“Would you pay $99 a month for a tool that solves X?”) rather than questions grounded in past behaviour (“Tell me about the last time X happened — what did you do, how long did it take, what did it cost you?”). ¹⁴

The distinction matters because you cannot trust what customers say they will do in the future. You can only trust how they currently do things. ¹⁵ Hypothetical commitment is not commitment. It is politeness in future tense.

Selection bias compounds the problem further. With typical sample sizes of five to fifteen customers, interviews cannot tell you whether a pain point is shared by five percent or ninety-five percent of the market. ¹⁶ Founders interview customers they have access to, not a representative sample. ¹⁶ They are also subject to recency bias: customers describe their most recent frustration, not their most important unmet need. ¹⁶ And when asked what they want, customers describe solutions — usually incremental improvements to what they already have. ¹⁶

The cost: forty-three percent of startups, one avoidable mistake

The downstream consequence of corrupted discovery is not abstract. CB Insights analysed 431 failed VC-backed companies that shut down since 2023 and found that 43% failed due to poor product-market fit. ¹⁷ This updated their earlier analysis of 110+ startup post-mortems, which had placed “no market need” at 42% — the same finding, four times the data, a decade apart. ¹⁷ The 431 companies in the study had raised a combined $17.5 billion in equity before dying. ¹⁸ What they lacked was not capital. It was evidence that anyone wanted what they were building. ¹⁸

In nearly every documented case, the information needed to avoid the failure existed before the product was built. ¹⁹ Target customers could have told the founders that the problem was not painful enough, that existing solutions were adequate, or that they would not pay the target price. ¹⁹ The founders either did not ask, or asked in ways that produced misleading answers. ¹⁹

Steve Blank’s customer development methodology, which launched the lean startup movement, was built on precisely this insight: there are no facts inside your building, so get outside to test them. ²⁰ The hypothesis-testing process emulates the scientific method — pose a business model hypothesis, design an experiment, get out of the building and test it. ²⁰ But “getting out of the building” is only valuable if the questions asked outside it are designed to falsify, not to confirm.

A protocol for falsification-first discovery

The antidote to confirmation bias in discovery is not greater scepticism in the moment. It is a workflow designed before the first interview begins — one that pre-commits to success criteria, captures interviews verbatim, scores evidence against a rubric, and forces disconfirming evidence into the decision. ²¹

First: pre-register your kill conditions. Before any interview, write down the specific finding that would cause you to abandon or substantially revise the hypothesis. This is not pessimism; it is the scientific method applied to product decisions. If a question will not change your business direction regardless of the answer, it is not worth asking. ²² Pre-registration prevents the most common post-interview failure: reinterpreting what you heard to fit what you hoped.

Second: ask about the past, not the future. The Mom Test’s core principle is that all questions should be about the customer’s actual life and past behaviour — not your idea or their hypothetical opinions. ²³ Ask about specific past events, not generics. Talk less and listen more. Do not ask whether they would hypothetically use your product; ask how they currently address the problem your product would solve. ²⁴

Third: reach beyond your network. The people with whom you have close personal relationships may lack objectivity or worry about offending you. ²⁵ The more interviewees realise they are evaluating a founder’s idea, the more social desirability bias distorts their answers. Distance yourself from the company during recruitment; where possible, have a neutral party conduct the interviews. ²⁶ Interviewers should avoid giving verbal or non-verbal cues that might influence responses and use neutral reactions to encourage more honest answers. ²⁶

Fourth: run a pre-mortem before synthesis. Organisational psychologist Gary Klein formalised the pre-mortem technique in a 2007 Harvard Business Review article: before finalising any decision, the team imagines the project has already definitively failed and works backward to identify why. ²⁷ Research on prospective hindsight — imagining an event has already occurred rather than imagining it might occur — increases the ability to correctly identify failure reasons by 30%. ²⁸ Applied to discovery synthesis, a pre-mortem forces the team to articulate the disconfirming signals they observed but were tempted to dismiss. Nobel laureates Daniel Kahneman and Richard Thaler have both encouraged its use. ²⁹

Fifth: validate with behaviour, not words. Customers may say they want or need something, but they may not actually use it or pay for it when it is available. ³⁰ Experiments that test customers’ actions — landing pages, prototypes, minimum viable products, pre-orders — measure willingness to act, not willingness to agree. ³⁰ The Sean Ellis test operationalises this at a later stage: ask active users “How would you feel if you could no longer use this product?” — if 40% or more say “very disappointed,” you have genuine signal. ³¹ Below that threshold, you are building on hope. ³¹

What this means

The decision discipline that discovery demands

Product discovery is not a feed of customer sentiment. It is a decision-making process with a specific output: a belief, either confirmed or killed. Every interview that ends without updating a hypothesis in either direction has produced noise, not signal. The founder who walks out of ten interviews with the same beliefs they walked in with has not done discovery — they have done reassurance.

The cognitive science here is unambiguous. Confirmation bias acts at the search, interpretation, and memory stages simultaneously. ³ Social desirability bias ensures that customers, in the presence of an enthusiastic founder, will systematically underreport scepticism. ⁷ Leading questions prime the answers before the customer opens their mouth. ¹² And a sample of five to fifteen interviews, drawn from the founder’s own network, cannot distinguish a niche preference from a market need. ¹⁶

None of this is a counsel of despair. The customer interview remains the most powerful qualitative instrument available to a founder — precisely because, when designed correctly, it can surface the uncomfortable truth that saves a company from building the wrong thing. The discipline required is not more interviews. It is better ones: falsification-first, past-behaviour-anchored, verbatim-captured, and pre-mortem-reviewed. That is what separates discovery from self-deception.

At Business Growth Accelerator (a FounderWise brand), the structured discovery protocol — pre-registered hypotheses, kill conditions, and post-synthesis pre-mortems — is embedded in how founders approach their earliest market conversations. The goal is not validation. It is truth.