Why unstructured interviews fail
The standard interview is a conversation that ranges wherever the interviewer takes it, ending with an impression. That impression correlates strongly with how much the interviewer enjoyed the conversation and only weakly with whether the person can do the job.
Three specific failures follow.
Candidates are not comparable. Asked different questions, they cannot be scored against each other, so the decision falls back to which conversation felt best.
Confirmation takes over fast. A judgement forms in the first few minutes, and the rest of the interview is spent gathering support for it rather than testing it.
Similarity wins. Without criteria, "good fit" reliably means someone with a similar background who communicates the way the interviewer does.
The minimum viable structure
Criteria written before you meet anyone, the same core questions for every candidate, and scores submitted independently before anyone discusses. Three things, none of which require an HR department, and together they do most of the work.
The four stages
Screen, 20 minutes
Confirm the basics: what they are looking for, whether the pay range works, availability, and that they understand the role as described.
Its only job is to prevent both sides wasting a full interview on a mismatch that a five-minute conversation would have surfaced.
20 minutesFilters mismatches, not candidatesCore interview, 60 minutes
The same core questions for everyone, focused on what they have actually done, with follow-ups on their reasoning.
Two interviewers where possible, scoring independently. One person's read of a conversation is a single, unverified data point.
60 minutesSame questions, every candidateWork sample, capped at two hours
Something resembling real work, sized so that a person with a job and a life can complete it. Pay for it if it is substantial.
This stage predicts performance better than any interview question, because it observes the thing rather than a description of it.
Two hours maximumTeam conversation, 45 minutes
The people they will work with daily, focused on how they collaborate and what they would want to know about working here.
Also the candidate's chance to interrogate you honestly. Someone who asks nothing difficult is evaluating a job rather than choosing a company.
45 minutesBoth directions
Tell candidates the whole process up front, including timelines. A clear process is itself a signal about how the company operates, and it is one of the few things that distinguishes a small company competing against larger ones.
Questions that predict performance
Anchor everything in what they actually did. Hypotheticals measure imagination.
| Question | What it reveals |
|---|---|
| "Walk me through how you did X in your last role" | Actual involvement versus proximity to it |
| "What did you get wrong, and what changed after?" | Self-awareness, and whether they learn |
| "Tell me about a decision you made without enough information" | Judgement under uncertainty |
| "What did you disagree with, and what did you do?" | Whether they raise problems or absorb them |
| "What part of your last job did you do that was not your job?" | Range, and initiative |
| "What would you need from us in the first month?" | Self-knowledge, and whether they will ask for help |
Then follow up relentlessly on specifics. "Who else was involved?", "what did you personally do?", "what happened next?". People who genuinely did the thing answer easily and in detail. People describing something they observed become vaguer under follow-up, which is the most reliable signal in the whole interview.
The work sample
The strongest predictor in the process, and the one small companies most often skip.
Make it resemble the actual job. A sales role gets a real objection to handle. An operations role gets a messy situation to sort out. Abstract puzzles measure puzzle-solving.
Cap it at two hours and say so. Longer exercises quietly filter for people with free evenings, which is not a quality you are hiring for.
Pay for anything substantial, or use something you will not ship. Asking for free work that solves a real problem is a bad look and gets discussed publicly.
Score it against the same criteria you wrote at the start, not against a general impression of polish.
Deciding without arguing
Write the criteria before meeting anyone. Four or five, drawn from the job description. If you cannot write them, the role is not defined well enough to interview for.
Score independently, then discuss. Everyone submits before hearing anyone else. The moment one person says "I loved them", independent judgement is gone.
Discuss disagreements, not averages. Where two interviewers diverge sharply is where the useful information sits. Averaging the scores discards exactly the signal you convened to find.
Decide on evidence you can point to. "I am concerned because they could not describe their own role in the project they led" is evidence. "Something felt off" is a feeling that may be a real signal or may be bias, and it cannot be examined.
Then actually check references. Most small companies skip this or treat it as a formality, and it is one of the highest signal-per-minute steps available. Our guide to checking references properly covers how to get past the confirmation-of-dates conversation.
Everything here assumes you know what you are hiring for. If the role is still fuzzy, the process will produce a confident decision about the wrong thing, which is why writing the job description properly comes first.
Frequently asked questions
- How many interview stages should there be?
- Three or four for most roles at a small company. Beyond that you lose good candidates to faster processes, and each additional stage adds less information than the one before it.
- Should every candidate get the same questions?
- The same core questions, yes, with room to follow up on their specific answers. Comparing candidates who were asked different things is comparing impressions rather than evidence.
- Are take-home exercises fair?
- Only if they are short, paid or capped at about two hours, and resemble real work. Long unpaid exercises exclude people with caring responsibilities or second jobs, which narrows your pool in a direction you did not choose.
- What is a good interview question?
- One about something the candidate actually did, with follow-ups about the reasoning. Hypotheticals measure how well someone imagines, which is a different skill from doing.
- How do you avoid hiring people just like you?
- Score against criteria written before meeting anyone, and have each interviewer submit scores before discussing. Both counter the pull toward candidates who are enjoyable to talk to, which usually means similar to you.