Home/Blog/Early Hiring/Run a Work Sample Test: Real, Paid and S…
Early Hiring

Run a Work Sample Test: Real, Paid and Scored Fairly

How to run a work sample test: set a short, real task, pay for the time, ask everyone the same questions and score against your ninety-day document, not a mood.

Cover of the FIKR Space guide: Run a Work Sample Test: Real, Paid and Scored Fairly

To run a work sample test, give each finalist a short, genuine task that resembles the real job, pay them properly for their time and ask the same questions of everyone. Watching someone do a small version of the work is the most reliable predictor of whether they can do the job.

Interviews are weak predictors of performance and everybody knows it, yet most early hiring is nothing but interviews. A modest amount of structure does far more than a large amount of intuition, and the work sample is the piece of structure with the greatest effect.

How to run a work sample test from start to finish

The test has four parts: the task, the payment, the questions and the scoring. Each is simple. The value comes from doing all four rather than the one that is easiest.

PartWhat to doWhy
The taskA few hours, resembling the actual workIt shows how they work, not how they talk
The paymentPay them properly for their timeIt makes the request fair and attracts serious people
The questionsThe same set for every candidateIt makes comparison possible and bias visible
The scoringAgainst the ninety-day document, written down as you goMemory drifts towards whoever you liked

Treat the table as a minimum. If you can do only part of it, do the parts that protect fairness: the same questions and the written scoring.

Choose a task that is genuinely the job

Design a short, genuine task. A few hours is the right scale. It should resemble the real work, and ideally it should be something you would have had to do anyway. That last point keeps the test honest, since the exercise is then a small piece of real work and not an artificial puzzle.

The task should come from the role, so go back to your outcome document. The page you wrote in how to write a ninety day role outcome tells you what the person will be asked to produce. A good test is a small slice of that.

Free guide, 33 pages: The First Ten Hires Playbook. Send me the guide.

Leave the brief a little incomplete

You are not only checking capability. You are watching how someone thinks, how they handle ambiguity, what questions they ask and what they do when the brief is incomplete. None of that is visible in a conversation, and all of it predicts the next year better than any answer to a hypothetical.

That has a practical consequence. If you polish the brief until nothing is ambiguous, you hide the very behaviour you want to see. Real work in a young company arrives with gaps, and the test should too. Decide beforehand which gaps are acceptable, and note how each candidate responds to them.

Pay for it properly

A work sample should be paid, and paid properly for the time it takes. Asking for free work from people who may have several options is a poor start to a relationship, and the candidates most worth having are the ones with the most alternatives.

Payment also disciplines you. If you are paying for the time, you will make the task short and relevant, and you will think harder about what you want to learn from it. It signals respect for the candidate, and respect is part of the selling that early hiring requires. The honest-selling side is discussed in how to hire fast as a startup.

Ask the same questions of everyone

Unstructured interviews mostly measure how much the interviewer enjoyed the conversation, which correlates with similarity and not with performance. A short, consistent set of questions asked of every candidate makes comparison possible and makes bias visible.

Use the review of the task as one of those occasions. Ask each candidate to walk you through what they did, what they chose to leave out and what they would do with more time. Keep the prompts identical. When the questions are the same, differences in the answers mean something.

Write notes while it happens

Write down the answers as you go, not afterwards. Memory reliably reshapes itself towards whoever you liked, so notes made later are partly a record of your mood. Notes made during the conversation are a record of what was said.

Do not leave the notes for later in the day either. The gap between the conversation and the write-up is exactly where the reshaping happens, so the habit of writing during is the whole point.

Score against the document, not the impression

Score each candidate against the ninety-day outcome document rather than against an overall impression. The question is not whether you liked them. It is whether what they produced and how they behaved points towards the outcome you described.

This also connects to the bar you set earlier. You wrote what a clear yes and a clear no look like before meeting anyone. The score is where you apply it. If the result is ambiguous, the rule from earlier stands: ambiguity means no, because a hire you are unsure about is one you will be unsure about for a year.

A simple scoring sheet

ItemWhat you record
The outcome lineWhich part of the ninety-day document the task relates to
What they producedYour notes on the work itself
Questions they askedWhat they chose to clarify, and when
Handling of gapsWhat they did when the brief was incomplete
Answers to the shared questionsThe same prompts, answered in their own words
Against the barClear yes, clear no or ambiguous

The sheet is plain on purpose. Its job is to make sure each candidate is described in the same terms, so the comparison is between evidence and not between moods.

FIKR People, team and HR for founders: hiring, onboarding and reviews in one workspace. See FIKR People.

What the test tells you that talking cannot

A conversation can describe how someone approaches a problem. A small piece of real work shows it. The difference matters most in the places a hypothetical question cannot reach: what a person does first, what they decide not to do, and how they behave when something is missing.

Consider the questions a candidate asks. Someone who asks nothing may be confident, or may not have noticed the gap. Someone who asks a sharp question early has shown you a habit you will be working with every day. Neither reading is certain, which is why the questions are recorded and compared across candidates and not judged in isolation.

After the test: follow up what came up twice

When the same point appears more than once, in the task, in the answers or later in a conversation with someone who knows the person, follow it up instead of explaining it away. A single oddity can be noise. Something that repeats deserves a direct question.

The natural next step is verification from outside. The work sample shows what a person does with you for a few hours. The people who have worked alongside them over a long period can tell you what the sample cannot, and how to do reference checks covers how to ask them well.

Keeping the process fast

A paid task adds a step, so plan for it. Set the dates for the task and the review in advance, tell candidates the whole timeline, and keep the gap between the task and your answer short. A strong candidate who is left waiting while you compare others may take another offer, and speed is one of the few advantages you have.

Why this beats another round of interviews

A short paid work sample predicts the next year far better, and it takes less of everyone's time than a fourth round of conversations. For the candidate, a few hours of real work is a more respectful use of their time than repeated meetings that cover the same ground. For you, it replaces a pile of impressions with something you can look at.

It also feeds your later learning. After a hire, you can ask whether the test surfaced what mattered. That review is part of building each hire on evidence from the last, which how to tell a hire is not working returns to.

Free guide, 33 pages: The First Ten Hires Playbook. Send me the guide.

Mistakes with work samples

  • Making it unpaid. It excludes serious candidates and starts the relationship badly.
  • Making it long. A few hours is the scale. Anything bigger stops being a sample.
  • Making it artificial. A puzzle unrelated to the work tells you about puzzles.
  • Changing the questions between candidates. Comparison depends on consistency.
  • Scoring afterwards from memory. Write as you go, against the document.
FIKR People, team and HR for founders: hiring, onboarding and reviews in one workspace. See FIKR People.

Frequently asked questions

How long should a work sample take?

A few hours. It should be short enough to be fair to the candidate and real enough to resemble the actual work.

Why should it be paid?

Because it takes real time and the strongest candidates have options. Paying properly makes the request fair, and it pushes you to keep the task short and relevant.

What should I look at besides the finished work?

Look at how they think, how they handle ambiguity, what questions they ask and what they do when the brief is incomplete. None of that shows in a conversation.

What if the result is unclear?

Apply the rule you set in advance: ambiguity means no. If something came up twice, follow it up instead of explaining it away.

The full method, with the checklists, is in The First Ten Hires Playbook, a free 33-page playbook.

LG

Luis Goncalves

// Founder & CEO at FIKR Space

Three-time founder. Built and exited Evolution4All before this. Now building FIKR Space — the operating infrastructure underneath every innovation ecosystem (startups, accelerators, governments, investors). Lisbon-based, works global.

Claim your founding member spot.

Every founding member is hand-reviewed. Tell us who you are and how you want to co-create — we get back inside 48 hours.

Role
Job Title
Country
Cookies

We use cookies.

Essential ones keep the site working. The rest only run if you say so. No dark patterns, no pre-ticked boxes.Privacy policy

Co-create with us. It's about building it together.Co-Create With Us