Interview process for startups

A good interview process is fair, fast, and consistent: every candidate meets a similar loop, gets judged on the same criteria, and hears back quickly. You do not need a hiring committee or a psychometric battery to get there. You need a short structured loop, a scorecard, and the discipline to test the actual work instead of trusting a good conversation.

This guide covers the loop, the scorecards, who to involve, how to keep bias out, how to respect the candidate’s time, and how to move a hire cleanly into onboarding.

Structure beats instinct

The single biggest improvement most startups can make is to run the same interview loop for every candidate. Unstructured interviews, where each conversation goes wherever it goes, feel insightful and predict almost nothing. They mostly measure how much the interviewer liked the person, which is rapport, not ability.

A structured loop for a small team looks like this:

1. A short screen. Fifteen to thirty minutes to confirm the basics, the level, and mutual interest. Cheap to run, saves everyone a full loop that was never going to work.
2. One or two focused interviews. Each tests something real: a work sample, a problem close to the actual job, a walk through past work in depth. Not a quiz, a rehearsal of the role.
3. A final conversation. Usually with a founder or the hiring manager, covering scope, expectations, and the candidate’s own questions.

Three or four stages is the right size. More than that and you are exhausting people and slowing yourself down without learning more. Each stage should have a stated purpose, so you are testing something different each time rather than asking four versions of the same question.

Scorecards, and why they matter

For each interview, decide in advance what it is testing and what good looks like. Write it down as a small set of criteria. After the interview, every interviewer scores those criteria and records the evidence: what the candidate actually said or did, not a gut verdict.

Two rules make this work.

• Define the criteria before you meet anyone. Criteria written after the fact bend to fit the person you liked.
• Score independently, then discuss. Interviewers submit their scores before the debrief. Otherwise the loudest or most senior voice sets the room and you get agreement instead of signal.

Scorecards do three things at once. They give you independent evidence rather than a consensus formed in the first thirty seconds. They make the process consistent between candidates, which is what fairness actually is. And they leave a written record, so a decision can be explained later to the candidate, the team, or yourself.

Who should be involved

Keep the panel small and deliberate. Every interviewer should have a defined thing they are assessing, so you are not running the same conversation several times.

A workable split for a startup: the hiring manager owns the role and the final call, one or two people assess the craft (they can judge the actual work), and a founder covers scope and the two-way fit. That is enough. Adding interviewers past the point where each has a distinct job dilutes the signal and burns the candidate’s day.

Brief the panel before the loop. Everyone should know the outcome the role owns, what their stage is testing, and what the others are covering. An uncoordinated panel double-tests some things and misses others.

Keeping bias out

Bias in hiring is rarely anyone being malicious. It is the accumulation of small unstructured judgements, each one slightly favouring people who remind the interviewer of themselves. Structure is the antidote.

• Ask every candidate the same core questions, so you are comparing like with like.
• Score defined criteria, not a vague sense of fit. “Culture fit” tends to mean “similar to us”, which is how a team accidentally hires one type of person over and over. Replace it with specific values and behaviours you can point to in the evidence.
• Focus on evidence of doing the work: a sample, a real problem, a detailed walk through past projects. Weight that over credentials and rapport.
• Have interviewers record scores independently before comparing notes.

And the plain one: no AI should rank or score candidates for you. A model trained on your past hiring learns your past bias and applies it at scale behind a neutral-looking interface. The rejected candidate never learns why, you often cannot either, and as the employer you carry the liability for a biased outcome. A human should read every application and run every interview. Capstan does no AI screening or ranking of candidates anywhere in the product.

Respect the candidate’s time

The best candidates have options and a job already. A slow, sprawling process loses them, and it tells them how you will treat them once they are inside.

• Keep the loop inside two weeks where you can. Momentum is a signal in both directions.
• Tell people the stages up front, so they can plan.
• Batch the interviews rather than dragging one a week across a month.
• Reply, always, including to the people you reject. A prompt, human no costs you a minute and earns you a reputation.
• Do not ask for unpaid work that mostly benefits you. A focused sample is fair; building your feature for free is not.

Fast and respectful is not just kindness. It is how you win the candidate who is also interviewing elsewhere.

What a tool should and should not do here

Interviewing is the part of hiring where software most often oversteps, so it is worth naming what help actually looks like.

Useful: an interview recorded against the application with its panel and its time, and one scorecard per interviewer that cannot be edited after it is submitted. That last property is the one that makes independent scoring real rather than aspirational, because a scorecard you can quietly revise after the debrief is just a record of the group’s opinion. Also useful: a stage vocabulary you can extend, transitions that are audited, and a mandatory reason on a rejection, so a decision six months old still has its reasoning attached.

Worth knowing rather than discovering: Capstan’s Recruitment module records interviews and panels but does not sync calendars, so scheduling stays in whatever you already use. It stores a résumé and does not parse it into fields. Candidates do not get a login; they follow a signed, expiring link that shows their own application status and nothing about anyone else’s. And it does no AI screening, scoring or ranking, anywhere, which is a product decision rather than a gap.

Move the hire cleanly into onboarding

The moment a candidate accepts, the process should not fall off a cliff into a spreadsheet. The clean version carries the person straight from pipeline to record. In Capstan an accepted offer converts the candidate into an employee through the same onboarding machinery a directly added hire runs through, rather than a parallel path that behaves slightly differently, and the created person keeps a link back to the candidate they came from. So the name, start date, and documents you already have are not re-keyed, and onboarding starts from real data.

From there the onboarding checklist picks up: the tasks that assign themselves when a start date is set, so the person’s first day is ready before they arrive.

Where to go next

The loop tests against the role you defined in the hiring plan and advertised in the job description, so those come first. Once someone accepts, the onboarding checklist turns the offer into a good first day. You can also look at the product directly, where the core is free up to twenty active employees and the module prices are public.