Unstructured interviews don't work

Three decades of meta-analyses converge on the same finding: unstructured interviews — "let's just have a chat and see how it goes" — predict job performance about as well as flipping a coin. Structured interviews, with the same questions and the same rubric for every candidate, predict performance roughly four times better.

If you take only one thing from this post: write down your questions and your rubric before you talk to the first candidate, and use the same ones for everyone.

The four-stage structure

For most SMB roles, four conversations is the right amount of process. More than that, you lose candidates to faster competitors. Less than that, you make decisions on too little signal.

1. Screening call (15 min) — confirm interest, salary, start date. Filter out non-fits.
2. Hiring-manager interview (45–60 min) — work history dive-in.
3. Technical / craft interview (45–60 min) — can they do the actual work?
4. Final round with the founder or CEO (30 min) — culture, motivation, mutual sell.

Stage 1: the screening call

Three questions, no exceptions:

1. "Why this role, and why now?"
2. "What's your salary expectation? We have a range of [X–Y] for this position."
3. "What's your earliest start date?"

If any of those answers is a clear non-fit, end the call kindly. Do not move to stage 2 hoping the answer will change.

Stage 2: the hiring-manager interview

The structure that gets the most signal in 45 minutes:

• 5 min introductions and role recap.
• 25 min structured dive-in on 2–3 things from their CV. Pick the most recent role, the most relevant role, and one outlier project. For each, ask:
  • "What was the business problem?"
  • "What did you do, specifically?"
  • "What was the outcome, in numbers?"
  • "What would you do differently?"
• 10 min the candidate asks you questions.
• 5 min logistics — next steps, timeline.

The key word is you. Most candidates will say "we did X" on autopilot; your job is to disambiguate what they personally contributed.

Stage 3: the craft interview

The interview that tests whether they can do the work, in the format closest to the actual work. For each role type:

• Software engineer — pair-programming on a 30-minute problem rooted in your real codebase, not a LeetCode puzzle.
• Marketing / growth — written exercise: "Here's our homepage. Write a one-page critique with three concrete experiments to run in the first month."
• Sales — role-play a discovery call with you as the prospect.
• Product manager — "Walk me through the spec you'd write for [actual feature on our roadmap]."

Take-home assignments work in theory but cost the candidate 4–6 hours and have a 30% completion rate. Use them only if your shortlist is too long to interview live.

Stage 4: the final round

This is where the founder or CEO sells the company while also testing motivation. Three open-ended questions:

1. "Why this company, specifically? What about us got you to apply?"
2. "What's a team or environment you've worked in that brought out your best work? Describe it."
3. "What would make this not the right role for you, six months in?"

The third is the most important. Anyone who cannot articulate a real failure mode is either inexperienced or not being honest.

The rubric

For every interview, score the candidate on the same dimensions, on a 1–5 scale, with notes:

• Skills relevance (1 = no overlap, 5 = perfect fit)
• Demonstrated outcomes (1 = vague claims, 5 = specific shipped work with numbers)
• Communication (1 = unclear, 5 = crisp and structured)
• Self-awareness (1 = none, 5 = names own failure modes accurately)
• Motivation (1 = generic, 5 = specific reasons this company at this stage)

Total score, then a one-sentence recommendation. Do this before you compare notes with anyone else on the panel; otherwise the loudest voice wins.

Bias-checking

Three habits cut interview bias more than any "unconscious bias training" you can buy:

1. Same questions, same order, same length for every candidate.
2. Score before you discuss with the panel.
3. Audit your hires. Look at the demographic split of your applicant pool, your interview pool, and your hires. If they diverge sharply, something in the process is filtering on something other than ability.

Questions to never ask

For legal and decency reasons:

• Marital status, plans for children, partner's employment.
• Age, date of graduation as a proxy for age.
• National origin, religion, sexual orientation.
• Disability status (you can ask whether they need accommodations to perform the role; you cannot ask about diagnoses).
• Past salary in jurisdictions where it is illegal (most of the US, the UK, the EU under the Pay Transparency Directive).

If you genuinely don't know whether a question is OK, the safe rule is: would I ask this of every candidate, regardless of demographics? If yes, fine. If no, skip.

Where Penroll fits

Penroll's CV ranking suggests interview questions tailored to each candidate's CV — concrete projects to dive into, gaps to ask about. You still run the structured interview yourself; the AI just gives you a better starting point than "tell me about yourself."