Same questions, work samples over CVs, and a decision made against the definition rather than against a feeling.
12 min read · Level 3
Unstructured interviews mostly measure comfort in interviews, which is rarely the job — and inconsistency makes comparison impossible while letting impressions substitute for evidence. "Culture fit" is the most common route to this failure, because undefined it reliably means "similar to me", which narrows the team and misses good people while feeling like judgement.
A structured process: the same core questions for every candidate, a realistic work sample, references, and a decision scored against the role definition.
The alternative — an unstructured chat followed by a gut call — is both less accurate and less fair.
Because unstructured interviews predict performance poorly. They mostly measure how comfortable someone is in an interview, which is rarely the job.
And because inconsistency is unfair in a way that compounds. Asking different candidates different questions makes comparison impossible and lets impressions substitute for evidence.
A structured process for the growth role:
1. Screen (20 min). Same five questions for everyone. Mostly: tell me about a channel you took from zero.
2. Work sample (90 min, paid). 'Here is our real product and our real ICP. Which channel would you test first, what would you spend, and what would make you stop?' This is the actual job, and it is far more predictive than any interview question.
⚠ Pay for anything substantial. Asking for hours of unpaid work is exploitative, and the best candidates decline.
3. Deep dive (60 min). Go through the sample. How did they think? What did they miss? How do they respond when you disagree?
4. References (2 × 20 min). Ask specifics: what did they own? what would you have them do differently?
5. Decision. Score against the role definition, written down before anyone advocates verbally. Feelings after the fact reshape memory.
Candidates who did not get the job get a real answer, promptly. It costs ten minutes and it is the difference between a person who speaks well of you and one who does not.
First-time founders often hire on 'culture fit', which frequently means 'similar to me'. That narrows the team and misses good people. Define the values you actually want and assess against those instead.
The second mistake: skipping references because they feel like a formality. Asking two specific questions — what did they own, what would you have them do differently — routinely surfaces things three interviews did not.
The third: ghosting candidates. Everyone gets an answer, even a short one. Startups have small reputations and long memories.
Same core set for every candidate, derived from the role definition.
Something resembling the actual job. It predicts far better than any interview, and paying for it is the minimum.
Walk through their reasoning. What did they consider and reject? How do they handle disagreement?
'What did they own?' and 'what would you have them do differently?' Both produce real answers.
Before discussion. Verbal advocacy reshapes everyone's memory of the interview.
Promptly, and with something real for anyone who reached a later stage.
Every hire, from the first one.
For someone you have worked closely with for years, a full process is theatre — you already have far better evidence than an interview would give you.
A realistic paid exercise predicts performance far better than any interview, because it is the actual job. Asking for hours of unpaid work is exploitative and, practically, the strongest candidates decline — so you lose the people you most wanted to assess.
Keep it to a couple of hours, make it resemble the real work, and pay for it. Then discuss their reasoning, not just the output.
Verbal advocacy reshapes everyone's memory of an interview — one confident voice in a debrief reliably shifts what other people believe they observed. Independent written scores against the role definition preserve what each person actually saw.
Everyone scores privately against the definition before anyone speaks. It is a small procedural change with a large effect on accuracy.
Create a free account to read the rest and apply it to your own startup.