Hiring salespeople: let the work sample decide

Sales recruiting carries a built-in trick that no other hiring does: the candidate is professionally good at exactly what the interview measures. A developer who interviews well still cannot code, and a code test reveals it. A salesperson who interviews well has just demonstrated parts of the job, and the recruiter walks away convinced, in precisely the way the candidate is trained to convince. The references? Hand-picked. The resume? Quotas without context. The only thing that cannot be curated is a conversation happening live.

Why does the interview fail specifically for salespeople?

Because the correlation between talking about the job and doing the job is weakest where talking is the job. LinkedIn's and Ipsos's study of 2,187 sellers across 15 countries found that only 18 percent qualify as what they call deep sellers, reps with the behaviors that make them nearly twice as likely to beat target. The behaviors in question, researching before contact, building relationships wide, prioritizing the right accounts, do not show in an interview, because in the interview everyone says they do them. The difference is only audible when the candidate has to do it in the moment: ask questions instead of pitching, absorb a no without losing footing, drive toward a next step without turning pushy.

How do you design the test itself?

Make it short, realistic and identical for everyone. Pick the conversation that dominates the role: a cold call for an SDR role, a follow-up or negotiation for a senior AE role. Ten to fifteen minutes is enough. Send a brief in advance with a product sheet and a short description of the counterpart, because preparation is part of the test: a candidate who shows up unprepared for an announced roleplay has also submitted an answer. Run the same scenario with the same counterpart for every candidate in the same role, otherwise you are comparing impressions instead of performances. And place the test early in the process, after the first screen but before the long interviews, so the hours go to candidates who can do what the job actually consists of.

The scoring rubric: six criteria are enough

Without a rubric the roleplay becomes just another interview, where overall impression wins. Score six things on a five-point scale: the opener, the discovery questions, the listening, the objection handling, the structure of the conversation and the drive toward a next step. Write one sentence of justification per criterion while memory is fresh, and have two people score independently before comparing. The spread between scorers is information in itself: criteria where you land far apart are usually vaguely defined. The point is not to reduce a person to numbers but to force specific observations, because 'strong candidate' is a feeling while 'lost the structure after the first objection but recovered the close' can be compared and followed up.

Standardize the counterpart

The test's weakest point is usually the scene partner. A sales manager playing the customer becomes unconsciously kinder to the candidate they already like, tougher on a tired Friday, and different with each person. One way to eliminate the variable is to let an AI play the customer: in Salesprep every candidate faces exactly the same persona with the same objections, the call is scored on six or seven components with a written comment on each, and afterwards you can compare the candidates' calls side by side instead of comparing memories. The candidate runs it in the browser with push-to-talk, and the raw audio is deleted after scoring, which keeps the setup easy to run fairly on the privacy side too.

The test's blind spots

A work sample measures skill in the moment, not stamina, learning speed or how the person behaves in November when the pipeline is thin. Two additions cover most of it. Ask structured questions about practice habits: how the candidate prepared for the test says something about how they will prepare for customer meetings, and Mindtickle's platform data suggests top performers practice roughly twice as much as the average. And separate current level from trajectory: a candidate who receives an instruction mid-roleplay and actually changes behavior on the spot is demonstrating coachability, which for junior roles can outweigh the starting level. Ask for exactly that in the test: give one concrete instruction halfway through and watch what happens.

How to run the process in a week

1. Define the role's most important conversation type and write a half-page scenario, with the same brief for every candidate.
2. Set the rubric: six criteria, a five-point scale, one sentence of justification per criterion and two independent scorers.
3. Screen on resume and a short call, and invite three to five candidates to the work sample instead of to a second interview.
4. Run the roleplays with an identical counterpart, ideally AI-standardized, and give a coaching instruction halfway through to measure receptiveness.
5. Compare the scores, take references with specific questions based on the test's findings, and spend the long interviews only on the top candidates.

Recruiting's old truth is that past results are the best predictor, but quotas from another company, another product and another market do not always travel. What travels is the behaviors, and behaviors can be observed in ten minutes if the situation is rigged right. Hire the person who sells in the test, not the person who sells in the interview.