Work sample tests: a complete guide for hiring teams

A work sample test is a small, real piece of the job that a candidate does before you hire them. A support applicant answers three real customer emails. An analyst finds what is wrong in a month of sales data. You judge the work, not the claims about the work. This guide covers when to use one, how to build and mark it, and what it looks like for six common roles.

What is a work sample test?

It is a hiring step with three parts:

• A task taken from the job. Something the person would really do in their first month, cut down to fit in an hour or less.
• The same task for every candidate, so you can compare them.
• Marking decided in advance, so the comparison is fair.

You will also hear it called a job simulation, a practical exercise or a job audition. A take-home assignment is a work sample done at home; a case interview is one done live. The idea is the same in each: watch the work, not the person describing it.

How well do work sample tests predict job performance?

Better than most things hiring teams rely on, and the honest answer has changed recently.

For twenty years the standard reference was a 1998 review by Schmidt and Hunter that ranked work samples at the top. In 2022 a team led by Paul Sackett re-ran the numbers with more careful statistics and published revised estimates in the Journal of Applied Psychology. On their scale, where higher means a closer link to later job performance:

Two things stand out. Work samples are still among the strongest methods, well ahead of the casual interview and far ahead of years of experience, which is what most résumé screening is really counting. And the method at the top, the structured interview, is the natural partner of a work sample: the best process uses the work as the subject of a structured conversation.

When should you use one?

Use a work sample when:

• The role produces something you can look at: code, analysis, writing, a campaign, a reply to a customer.
• You have more applicants than you can interview. A short task sorts a long list faster and more fairly than reading résumés.
• The applicants are early in their careers. A fresher has no track record to check, so the work is the only real evidence. India's hiring data makes the point: in HirePro's 2026 survey, only 9% of employers said fresh graduates are job-ready on day one.

Think twice when the job is mostly about relationships built over months, or when the person is so senior that their public record already shows the work.

How do you design a work sample test?

You can do it in an afternoon.

1. List what the person will do in their first ninety days. Pick the one task that best separates a good hire from a poor one.
2. Cut it to under an hour. Remove the setup, keep the judgement. If the real task is "clean this data, then find the problem", hand over data that is already half clean.
3. Use real, messy material. Tidy inputs test nothing. A date in the wrong format, two documents that disagree, a customer who is partly wrong: this is where skill shows.
4. Write the brief the way a manager would. Say what you need, by when, and what good looks like. Say which tools are allowed. Today that should include AI tools, because the job includes them.
5. Try it on someone on your team. If your own good people cannot finish it in the time, it is too long.

How do you mark it fairly?

Write the marking before the first applicant starts. For each thing you care about, describe what a weak, an acceptable and a strong answer looks like. Three to five criteria are enough: for example, "found the real problem", "checked their own work", "explained it clearly".

Then hold to two rules. Mark every candidate against the criteria, never against each other. And where you can, mark without the name and college in front of you, so the work is all you are judging.

Examples by role

Each one is a normal morning's work in that job. None is a puzzle.

What mistakes should you avoid?

• Making it too long. Past an hour or so for an early step, your strongest applicants leave first.
• Asking for free work. A task whose answer you could ship is not a test, and candidates can tell.
• Giving everyone identical material. Answers get passed around, and AI tools give every candidate the same result. Vary the material per person and the problem goes away.
• Judging the finish. Polish is free now. Read our piece on what to test once AI can finish the take-home.
• Skipping the conversation. A hand-in without a follow-up tells you half of what it could. See five follow-up questions about a candidate's own work.

Common questions

How long should a work sample test take? Thirty to sixty minutes when it comes early. Longer only for a final step, and then say so in advance.

Where in the process does it go? As early as you can. Used before interviews, it decides who to interview. Used after, it only confirms a choice you have mostly made.

Should candidates be allowed to use AI? Yes, if the job allows it. Design the task so that understanding is what gets marked.

Is it fair to candidates? Fairer than a résumé screen. It gives someone from a college you have not heard of the same chance as anyone else, provided the task is short, relevant and marked the same way for all.

Wooble runs this whole process for you: the task, each applicant's own material and the marking are already written for each role, and you get a ranked list with the reason for every place. See how it works.