Work samples and trials, without exploiting the candidate
A sample of the actual work predicts performance better than any interview. It also has an obvious failure mode that small employers fall into regularly.
If you can only add one thing to an interview process, add a work sample: a short, realistic task resembling something the person would do in the job, assessed against criteria set in advance.
It outperforms interviews as a predictor for most roles, and it has a second benefit that matters more in a small company — it gives the candidate a concrete picture of the work, which improves the quality of their own decision.
Realistic, small, and bounded
Three properties make a work sample useful and their absence makes it a burden.
- Realistic — drawn from work the role actually contains, not a puzzle. A bookkeeper reconciles a short messy statement; a support hire answers three awkward customer emails; a supervisor builds a week's rota from a set of constraints.
- Small — under two hours, and stated as a maximum rather than a suggestion. Open-ended tasks are completed by whoever has the most spare time, which selects for circumstance rather than ability.
- Bounded — clear brief, clear deliverable, clear deadline. An ambiguous task tests whether the candidate guessed your intent.
Decide what a strong, adequate and weak submission look like before you see any of them. A rubric written afterwards is a description of the submission you liked.
The line you should not cross
A work sample stops being a selection exercise and becomes unpaid labour at the point where the output is useful to you. The test is straightforward: would you use it, or could you?
Designing a real feature, drafting a real client proposal, producing a real marketing plan for the actual next quarter — these are work, and asking several candidates to do them for free is a poor practice with a real reputational cost. Small employers do this more often than large ones, usually without intending to, because inventing a fictional scenario takes more effort than using a live one.
Use anonymised, historical or invented material. If the task genuinely has to be live work, pay for it.
Paid trial days
For hands-on roles a paid trial — half a day or a day, working alongside the team, paid at the role's rate — is the most informative step available. Both sides learn a great deal that no interview surfaces. For broader independent guidance on this point, consult Wage and Hour Division.
The conditions that keep it fair are not complicated: pay it, be explicit that it is part of selection rather than employment, check the position on insurance and on what your local law says about trial work, and never run one so long that it becomes a way of getting work done cheaply. Two days is a trial; two weeks is a job.
Accessibility is part of the design
A timed task at nine on a Tuesday morning selects against anyone currently employed. A task requiring specialist software selects against anyone who does not already own it. A two-hour evening task is harder for someone with caring responsibilities. For a related product-level perspective on time-tracker manipulation, see more details.
Offer flexibility on when the task is done, supply whatever tools are needed, and be willing to adjust the format on request. This costs nothing and removes filters that have nothing to do with the work.
Give feedback on the sample
Candidates who spent two hours on a task have earned more than a form rejection. A few sentences on how the submission read against the criteria costs little and is remembered — including by people who will apply again, or who will mention you to someone who might.