The Paid Work-Sample Test That Beats Interviews
A paid work-sample test hands finalists a small, real piece of the job and pays them for their time. What they do with gaps and unclear facts shows how they'll work when you're away, which a conversation rarely reveals. It also costs far less than one hire who interviewed well and then couldn't decide anything alone.
Tuesday at 4:40 p.m., three folders sat on an owner's desk. Same packet in each: 20 scrubbed emails from a real week, a one-page note on how he handled things, and a request to sort them, draft the five most urgent replies, and list anything they would not touch without asking. Each finalist for the executive assistant seat had been paid $75. Two folders looked finished and confident. The third was thinner.
He almost threw the thin one out before reading it. The interview with that candidate had been flat. She had not sold herself. Then he opened the file. She had replied to 11 messages, flagged 6 as "needs you," and written one line about the other 3: "I could not tell whether this vendor is current or the one you stopped using in the spring. I held it."
Two of those three held a trap he had planted. A refund request that quoted a policy he had changed in January. A vendor invoice he had already disputed. The confident candidates had answered both. One promised the refund at the old terms. The quiet folder was the only one that refused to invent certainty.
That is what a paid work sample buys you. Not a polished product. A look at what someone does with a mess, with a gap, and with writing when nobody is watching.
Why do interviews pick the wrong hire?
I stopped trusting my own final interviews after watching an owner hire an operations coordinator in 40 minutes because "we clicked." She was funny. She had a warehouse move story. She had worked at three companies he knew. Ten weeks and $9,300 in wages later she had never once decided anything without asking him first, and he was answering about 25 Slack messages a day from her. The charm was real. The judgment never showed up in the conversation, because conversations do not ask you to hold a disputed invoice.
Charm is one of the ways owners end up hiring for relief; you can't hire your way out of owner dependency covers the rest.
A work sample is a 2 to 3 hour slice of the job, done on their time, paid. You pay because unpaid tests select for people with spare evenings and no other offers, and because paying turns it into a fair trade. For seats where the person will write for you by email or text, their writing is the hire. Charm in a chair does not equal that.
The quiet candidate in that 12-person home services firm did not win by being careful for its own sake. She won because she knew where her knowledge ended and said so in writing, which is the skill you need when you are on a plane and someone is about to invent a refund policy on your behalf. An interview rewards filling every silence; the seat needs someone who will leave a gap alone until they know the truth.
How do you build a work-sample test?
Start from the last two weeks of whatever the seat will touch. An inbox for an assistant. A scheduling board with three double-bookings for a coordinator. A month of unreconciled transactions for a bookkeeper. Scrub names and anything personal. Keep the ugliness. Messy is the point. If you invent a tidy exercise, every finalist will look the same and you will have spent $225 to learn nothing.
Plant three things and do not label them.
- One item with a right answer someone finds if they read closely.
- One ambiguous item where guessing is dangerous.
- One that looks urgent and is not.
Give a time box and a clear deliverable. Tell them questions are allowed in writing, and you will answer within a day. Whether they email you at all is data. One precise question at hour one is a working habit. Silence plus an invented answer is a different habit. Both habits show up again in month two, when you are gone and the inbox is live.
Score it blind. Strip names. Have the person who will sit next to this hire grade first, or grade yourself with names hidden, on four lines: did they find the right-answer item; what they did with the ambiguous one (asked, held, or guessed); whether the writing is something you would send under your name with no edits; whether they stayed in the time box or explained why not. A 1 to 5 on each is enough. You are making a tie-breaker that is not your gut.
One more design choice that matters: make the packet look like Tuesday, not like a test. If every message is labeled "Scenario A," candidates perform. If the packet looks like a forwarded inbox dump with your normal typos and half-finished threads, they work. The owner who got the 9-point spread used his real subject lines and left two threads mid-argument on purpose. Candidates who tidied those threads without asking had invented a resolution he did not want.
Owners resist two parts of this. Doing the final interview alone, and paying for a test. On the first: if you are the only filter, you have a long track record of being charmed. In that firm, the lead dispatcher scored the quiet candidate first and highest. The owner's gut had her third. He still finds that annoying, which is useful. On the second: three paid samples at $75 is $225 plus about 90 minutes of a lead's time to score. Put that next to $9,300 in wages on a bad hire, before you count the owner's cleanup hours, and the objection usually dies.
People who are good at the work often like a chance to show it. A finalist who bristles at a 2 hour, $75 task for a role paying around $55,000 a year is telling you something about ordinary Tuesdays. Not every decline is a problem. A field that all says yes and one that refuses is usually a clean read.
What went wrong with the first designs?
Our first sample was too clean. Twelve tidy tasks, clear instructions, and all three candidates scored within 2 points. We learned nothing. The second time we added the stale refund policy and a message that quietly contradicted the owner's own written note. The spread went from 2 points to 9. The traps did the teaching. The tidy checklist had only measured compliance.
Another owner paid $100, got excellent work, then learned a friend had helped. The dull fix works: a 10-minute follow-up where they walk through two of their choices. You are not grilling them. You are checking the reasoning belongs to them. Candidates who did the work can usually explain why they held the disputed invoice. Candidates who borrowed the work start talking in generalities about "being thorough."
The quiet hire was not a clean win either. She took 3 hours and 20 minutes against the 2 he asked for, and he almost marked her down for speed. In her first 30 days she asked him 41 questions, mostly in the first two weeks. By week six it was about 9 a week and the questions were about policy, not about where things lived. He still does not know how she would have done in a pure interview. He does know he almost skipped her for being quiet in a chair.
A paid sample is the wrong weight for some seats. A candidate for a $180,000 operations lead who has run a 60-person company may, after two strong conversations, answer your sample task by writing back, politely, that he does not do small paid tests and that you can call his former CEO. He is not being difficult. At that level a 2 hour exercise reads as a mismatch. Call three references who reported to him, not only people who liked him, then propose a paid 30-day trial at a day rate with a written scope and a checkpoint at day 14.
Skip the sample for skills you can train in a week, too: your specific software, your checklist. A trial shift tells you more about reliability than a homework packet. And if you have no mess to build from because the seat does not exist yet, stop. That is a role-definition problem, not a hiring test problem.
If the seat isn't defined yet, start with role scorecards instead of job descriptions and build the packet once the outcomes are written.
Did the quiet candidate work out?
He hired her. On her third Friday she held another invoice without asking, the same shape as the trap in the packet, and Slack stayed quiet until Monday. He still keeps the other two folders in a drawer, not as trophies but as a reminder that the candidate who finished every email was the one who would have promised the refund he no longer offered.
Frequently asked questions
How should I pay finalists for the sample?
Pay a flat amount for the time box, on the same terms for every finalist, and send it whether or not they get the job. Paying promptly through your normal contractor process keeps it clean and shows candidates how you treat people. Check with your accountant on how to record the payments where you operate.
What if only one finalist completes the sample?
Score it anyway against the same four lines, since a strong result still tells you a lot. A lone finisher can also mean the packet asked too much, so ask the others briefly why they passed. If their reasons point at the design, trim the packet before the next round instead of lowering the bar.
Can I reuse the same packet for the next hire?
Reuse the structure and refresh the contents. Candidates talk, and a packet that circulates loses its traps. Pull a new batch of real messages each time, keep the planted items doing the same jobs and leave the scoring lines unchanged, so finalists from different hiring rounds can be compared fairly.
What would your packet hold?
If you want help building the sample and rubric for a key seat, start here: systemstoscalegroup.com/getstarted
