Paying for the Take-Home Buys a Shorter Take-Home

Do you have to pay?

Usually not, under US federal law. Federal wage law reaches employees, and the Fair Labor Standards Act defines employ as to suffer or permit to work ¹. A candidate doing a short exercise whose output nobody uses generally sits outside that. The line is the use: the Supreme Court held in 1947 that trainees whose work gave the railroad no immediate advantage were not employees ².

Both of those need their limits stated. The 1947 case concerned a week-long practical course for prospective yard brakemen, and its modern descendant is the primary-beneficiary test courts apply to unpaid interns. Neither is a ruling about take-home assignments, state wage law can be stricter than the federal floor, and other countries answer this differently. This is public legal fact rather than advice, and an unpaid exercise that runs past a few hours is worth a conversation with counsel before it goes out.

The workable test is the one you can apply without a lawyer. If the deliverable would go to a customer, into your codebase, into a deck someone presents, or into a decision you were going to have to pay somebody to make, it is work. A candidate asked to audit your actual onboarding funnel and hand back recommendations is producing something you will use, whatever the stage is called. A candidate analysing a sanitised dataset that goes in the bin on Friday is not.

That test also catches the assignment nobody meant to write. Briefs drift toward live problems because live problems are easier to describe than invented ones, and the drift is exactly what moves an exercise from a test into unpaid consulting.

Who does an unpaid assignment filter out?

People whose evenings are already committed. Three unpaid hours is a fee payable in time, and the candidates least able to pay it are those with caring responsibilities, a second job, or a current role with a demanding week. None of that predicts how well somebody does the job. All of it predicts who returns the assignment, which is the part that reaches your shortlist.

Be honest about what kind of claim that is. It is a structural argument about who can afford an unfunded stage, not a measured finding, and no study here says that paying changes the composition of a specific pool. What makes it worth acting on anyway is the cost of being wrong in each direction. Paying costs a few hundred dollars a candidate. Not paying costs you the people you never see, and they leave no trace in any funnel report, because a candidate who quietly does not start is indistinguishable from one who was never interested.

Two design choices make the same exercise land very differently. An assignment placed after a human conversation asks for time from someone who now knows something about the role. An assignment used as a top-of-funnel filter asks for three hours from strangers, which is the version that produces both the worst completion rates and the loudest public complaints. Placing it late costs nothing and fixes most of this.

Disclosure does the rest of the work. When the job post names the assignment and its length, the people who cannot spend the time find out at the cheapest possible moment instead of after two rounds of interviews.

Pay a flat rate, and pay everyone at that stage

Flat rate, not hourly, and everyone who reaches the stage or nobody. A flat rate keeps the cap honest, because paying by the hour rewards whoever spends six of them. Paying only some candidates turns the payment into a signal about who you already favour, which is worse than paying none. Put the amount in the brief and in the job post.

The mechanics that actually derail this are administrative rather than ethical:

• Set the rate against the cap, not against the role. For a two-to-three hour exercise in a mid-level knowledge role, something in the range of a few hundred dollars reads as serious to candidates and rounds to nothing in a recruiting budget. Below about a hundred it reads as a gesture and occasionally as an insult.
• Agree the payment mechanism with finance before the round opens, once. This is where paid assignments die. Vendor onboarding for a $200 payment is absurd, so most teams settle on gift cards or a simple invoice, and the decision needs to exist before a candidate is waiting on it.
• Pay on submission, never on outcome. Payment conditional on passing is not payment, it is a prize, and it changes what candidates submit.
• Have an answer for people who cannot accept money. Current government employees, some contractors and some visa holders genuinely cannot. Offer an equivalent donation to a charity they name, record that the same rate was offered, and move on.

The rate does not move for the candidate you are most excited about. Raising it there is the same failure as paying selectively, wearing better clothes, and it shows the moment two candidates compare notes.

What changes once the hours are billed?

The assignment gets shorter, usually inside one cycle. Once the hours appear on an invoice, somebody asks what the fourth one is buying, and the honest answer is rarely something the second hour did not already produce. That conversation never happens while the cost sits entirely on the candidate, because a cost nobody in the room pays is a cost nobody in the room questions.

The budget objection is smaller than it feels. SHRM's benchmarking data puts median cost-per-hire for non-executive roles at $1,244, with a 25th percentile of $354 and a 75th of $4,375 against a mean of $4,683, and its definition sums agency fees, advertising, job boards, referrals, travel, relocation, recruiter pay and system costs ³. Two things follow. The distribution is heavily right-skewed, so no single number describes your situation, and the definition leaves out the hiring manager's and interviewers' hours, which are usually the largest real cost of a loop. Paying four or five candidates a few hundred dollars each sits inside the spread of what the process already costs, and it funds a stage that was previously financed entirely by people who did not get the job.

Three second-order effects worth expecting. You will send the assignment to fewer people, which is good, because reviewer time was always the binding constraint. Completion rates go up, which makes small differences between candidates easier to read. And the pressure to cap the exercise finally has somebody inside the company behind it, which is the only reason scope discipline ever survives contact with a hiring manager who wants one more question answered.

Paying does not make a long exercise acceptable, and it does not repair a vague one. It buys attention on the size of the thing. The related lever is the clock: what a time limit on an assessment actually measures is worth settling in the same sitting, since a paid cap and a graded timer are not the same instrument. And if the work is genuinely a week of somebody's life, you are not designing an assignment at all, which is the question behind running a paid trial project instead.

See what gets scored >