Why founders should not run their own user interviews

Founders should talk to users.

They should not always be the person asking the questions.

That sounds contradictory because early startup advice compresses two different jobs into one sentence. In the beginning, founders need direct exposure to customers. No summary, dashboard, or research report can replace the first-hand shock of hearing how people describe the problem in their own words.

But later, once the product exists and the next decision is expensive, founder-led interviews start to fail in a predictable way.

The call sounds warm. The user nods. The founder explains a little too much. The participant says the feature looks useful, the pricing makes sense, and the new workflow seems clearer. Everyone leaves feeling good.

A week later, the transcript contains almost nothing that would have changed the roadmap.

This is not because founders are careless. It is because the founder is the most invested person in the room. That changes what they ask, what they notice, what they remember, and what the participant feels comfortable saying.

The useful rule is not “founders should never interview users.” It is sharper than that:

Founders should run the early conversations that build understanding. They should step out of the moderator seat for the conversations that test their conviction.

The evidence is indirect, but strong enough to matter

There is not a perfect academic study that compares the same SaaS interview run by a founder, a neutral researcher, and an AI interviewer.

So the case has to be built from two bodies of evidence.

The first is survey methodology. Researchers have spent decades studying interviewer effects: the ways the person asking the question changes the answer. A Public Opinion Quarterly study on CATI, IVR, and web surveys found that interviewer-administered and self-administered modes produce different reporting patterns, especially around sensitive information. Pew Research Center’s work on mode effects shows the same broad point in public opinion research: the channel and social setting can shift answers, sometimes materially.

The second is user-research practice. Nielsen Norman Group’s User Interviews 101 names social-desirability bias and interviewer skill as core limitations of interviews. Their guidance on leading questions shows how easily wording, tone, and implied expectations nudge participants toward a particular answer.

Now add the founder.

A founder is not a neutral interviewer. The participant knows the founder built the product, cares about the answer, and may react emotionally to criticism even if they try not to. The founder knows what answer they hope is true. The social and cognitive pressures point in the same direction: softer participant answers, more leading founder prompts, and stronger memory for confirming evidence.

That is enough to treat founder-moderated interviews as a high-bias instrument for high-stakes product decisions.

The participant is managing you

When a customer talks to a founder, they are not only describing their experience. They are managing a relationship.

They may want to be helpful. They may not want to hurt your feelings. They may worry that criticism will turn into a debate. They may suspect that a blunt answer will make future support, pricing, or account conversations awkward.

Nielsen Norman Group’s article on the Hawthorne effect and observer bias is about a broader research problem: people change behaviour when they know they are being observed. In founder-led interviews, the observation is personal. The person watching is also the person whose work is being judged.

That is why the criticism gets softened:

• “I could not find it” becomes “I probably just missed it.”
• “We are evaluating a competitor” becomes “We are still figuring out the workflow.”
• “The price is not justified” becomes “Budget is a bit tight right now.”
• “I do not trust the output” becomes “I need to spend more time with it.”

Those are not lies. They are socially safer versions of the truth.

The founder is managing the evidence

The participant is only half the problem.

The founder also bends the interview, usually without noticing.

Teresa Torres’ piece on confirmation bias explains the practical trap for product teams: we notice evidence that supports what we already believe more readily than evidence that challenges it. In an interview, that bias is not abstract. It appears in the follow-up you ask, the awkward answer you move past, and the quote you remember afterwards.

Rob Fitzpatrick’s The Mom Test names the founder version of the same failure. Compliments, fluff, and ideas are bad data, and founders invite all three when they describe the idea too early or ask for opinions about the future.

Steve Blank’s site hosts Andrew Elliott’s post-mortem, How Customer Development Failed Us, which has the line every founder should read before a validation round: “the earlyvangelists you thought you had are actually just very polite users.” The painful part is that Elliott did not skip customer development. He did the work and still got false positives because the conversations were too easy to hear as validation.

Jeanette Mellinger makes this explicit in First Round Review: the more you have at stake, the more likely you are to listen for what you want to hear. She names confirmation bias, acquiescence bias, and hindsight bias as founder discovery risks. That triad is almost a description of a bad founder interview.

What founder bias looks like in the transcript

The signs are usually visible on the page.

The founder talks too early

The interview begins with context, then product background, then a short explanation of why the team is excited. By the time the participant answers the first real question, they already know what kind of answer would be encouraging.

A neutral interviewer delays the product. They ask about the participant’s world first.

The founder completes the thought

The participant says, “The reporting workflow is a bit...” and the founder jumps in: “Manual?” The participant nods. Now the transcript contains the founder’s word, not the participant’s.

The better follow-up is: “Tell me more about ‘a bit’. What happened?”

The founder validates the answer

Positive answers get warmth. Negative answers get careful reassurance. Both teach the participant what answers are emotionally loaded.

“That is great to hear” and “thanks for being honest” are both understandable. They are also signals. A skilled interviewer keeps the reaction smaller and asks for the next specific.

The founder defends in miniature

The participant describes confusion. The founder says, “That makes sense. We did change that flow recently.” It feels harmless, but it shifts the room from understanding to explanation.

The participant learns that criticism creates a response. The next criticism gets smaller.

The founder skips the painful follow-up

The participant says they would never pay the current price. The founder moves to the next topic.

That was the interview.

The useful question was: “What makes that feel too high? What were you comparing it with when you decided?”

When founders should run the interview

There is still a place for founder-led interviews.

Founders should run the first conversations when they are still learning the shape of the problem. Early discovery is not only about data collection. It is about building taste, empathy, vocabulary, and pattern recognition. The founder needs to hear how customers talk, where the story bends, and which assumptions sound ridiculous once spoken aloud.

Steve Portigal’s advice on bias in research is useful here. The goal is not to pretend you have no assumptions. The goal is to write them down, make them visible, and stop confusing them with evidence.

Founder-led interviews are also acceptable when the stakes are low and the goal is exposure: onboarding ride-alongs, support follow-ups, light customer check-ins, or early problem exploration where the founder is mostly listening.

The danger zone begins when the interview is supposed to validate a specific bet:

• Should we build this feature?
• Which segment should we target?
• Why did this customer churn?
• What would make this buyer switch?
• Is the price justified?
• Which message should lead the homepage?

Those conversations decide what you build, who you sell to, and what you charge. The founder has too much to lose from the wrong answer to be the cleanest instrument for collecting it.

What to do instead

You do not need to hire a research team before talking to users. You need to separate roles where it matters.

1. Put someone else in the moderator seat

Use a co-founder, teammate, advisor, freelance researcher, or trusted operator who did not build the thing being tested. Give them the research goal and a short guide. Then observe silently.

Erika Hall’s Interviewing Humans argues for separating interviewer, observer, and notetaker roles. Even if you cannot staff all three, the principle holds: the person asking questions should not also be defending the product, taking notes, and deciding what the answer means.

2. Use a guide that asks the hard questions

If you have to moderate, write the guide before you are in the call, and include the questions you are tempted to avoid.

Examples:

• “What nearly stopped you from buying?”
• “What would make you switch away?”
• “What were you doing before you tried us?”
• “What did you expect to happen that did not?”
• “If you had to cut one tool this quarter, where would we sit on the list?”

Do not ask all of them in every interview. But make sure the guide has room for the answer that would hurt.

3. Use a neutral interviewer for sensitive topics

Churn, pricing, win/loss, failed onboarding, and competitor evaluation are exactly where politeness costs you the most.

This is one of Maren’s strongest use cases. Participants talk to Maren instead of to the founder. She has no product to defend, no feelings to protect, and no reason to accept a vague answer. She can ask, “Can you walk me through what happened?” across many conversations and bring back the patterns.

That does not make AI a universal replacement for a skilled human researcher. It does make a neutral AI interviewer a practical option when the alternative is the founder accidentally running a sales call.

4. Read the transcript, not just the summary

Confirmation bias does a lot of damage between the conversation and the memory of the conversation.

Read the transcript later. Look for places where the participant hesitated, contradicted themselves, or gave a specific example you missed in the moment. Mark the exact words, not just your interpretation.

Indi Young’s Listening Deeply is a useful standard: the work is to follow the participant’s thinking, not your explanation of it.

5. Compare words with behaviour

Interviews are necessary. They are not sufficient.

NN/g’s User Interviews 101 makes the limitation explicit: if you want to know what users actually do, observe them or collect behavioural data. If a participant says onboarding was easy but the activation funnel collapses at step two, believe both signals enough to investigate the mismatch. If a churned customer says “budget” but usage had already dropped to zero six weeks earlier, the cancellation story started before the invoice.

The interview gives you meaning. Behavioural data gives you the check against self-report.

A practical rule

Use this split:

The pattern is simple. The more the answer can hurt the founder’s current belief, the less the founder should be the one asking.

The short version

Founders should talk to users early and often. That is still true.

But the founder is not a neutral research instrument. Participants soften criticism for founders. Founders notice confirming evidence. Leading questions appear in friendly clothing. Polite answers become roadmap conviction.

For early discovery, founder-led conversations are useful. For high-stakes validation, churn, pricing, win/loss, and roadmap decisions, step out of the moderator seat.

Have someone else ask. Use a guide that invites hard answers. Read the transcripts. Compare what people said with what they did. Use Maren when you need a neutral interviewer who can ask specific follow-ups at scale.

You still make the product decision. You just do not have to collect the evidence in the most biased room available.