The Reference Call Is Theater. Here's What Actually Predicts Performance.
Three curated references will tell you nothing you couldn't already guess. A paid work sample will.
I have made a lot of these calls. Head of sales at three companies, hiring probably 150 AEs and SDRs across fifteen years, and I would guess I've run 400-plus reference checks. I can tell you the modal outcome of a sales reference call in one sentence: the candidate is, according to three people the candidate personally selected, quite good. That is the finding. Every time.
This isn't a coincidence and it isn't because sales candidates happen to be uniformly excellent. It's because the reference check has a selection problem baked into its design that no amount of "asking better questions" fixes. You are asking a candidate to name three people who will vouch for them. Nobody hands you the manager they clashed with, the deal desk lead who caught them padding a forecast, or the SDR they burned out and pushed off the team. You get the friendly ex-boss, the peer who liked them, and a customer who remembers the good quarter. Three-for-three positive isn't evidence of quality. It's evidence the candidate can count to three.
The structural problem, not a process problem
Sales leaders tend to respond to this by trying to run a "better" reference call — sharper questions, asking for specifics, listening for hesitation. All useful skills. None of them touch the actual defect. You cannot out-interview a sampling bias. If the population you're drawing from was hand-selected to be positive, no amount of Colombo-style follow-up questions turns that into a representative signal. You will occasionally catch a lukewarm tone or a suspiciously short answer, and you'll feel like a detective. What you've actually detected is that this particular chosen advocate is a worse liar than the other two. That's not the same as finding out whether the candidate can run discovery.
Compare it to how you'd never accept this design anywhere else in the hiring process. You wouldn't let a candidate write their own performance review from a past job and submit it as evidence. The reference call is functionally that, laundered through a phone call with someone else's voice.
What actually produces signal
The fix isn't "better references." It's replacing the reference call, or at minimum de-weighting it hard, with a paid work sample: a real task, scored against a real rubric, done live or asynchronously by the candidate themselves, with money changing hands so it's a fair ask.
For a sales hire this is not hard to build. It looks like:
- A live discovery call role-play, 20 minutes, against a semi-hostile buyer persona built from your actual ICP — not a straw man who agrees with everything.
- A cold call segment, 5 minutes, cold open through to booking a next step, scored on structure not charisma.
- A written follow-up, 15 minutes, turning the call notes into next steps and a mutual action plan.
- A short debrief where the candidate is asked to self-assess against the rubric before you reveal your score — this alone tells you more about coachability than any reference ever will.
Pay for it. Ninety minutes at a fair day-rate, roughly £150-£250 depending on seniority and market. This is the detail sales leaders skip and it's the one that changes candidate behaviour most. An unpaid "trial" gets you people with time on their hands. A paid one gets you people who take the process seriously because you've signalled you take it seriously, and it removes the "we're just using you for free labour" objection that scares off your best candidates — the ones who are already employed and don't need to prove anything to anyone for free.
What you're actually scoring
This only works if the rubric is a competency rubric, not a vibe check. Score against named skills — discovery questioning depth, active listening, objection handling, qualification rigour — not "came across confident" or "good energy." If you don't already have named competencies to score against, that's the actual gap, and it's worth fixing before you fix the interview process, because a work sample scored on gut feel is just a more expensive version of the problem you're trying to solve. The Sales Candidate Scorecard: The 7-Competency Hiring Rubric gives you the structure to score the same session two different interviewers just watched and land within a point of each other — which a "how did that feel to you" debrief never will.
The comparison that matters
| Signal source | What it actually measures | Failure mode |
|---|---|---|
| Reference check | Candidate's ability to select 3 friendly contacts | Zero negative signal by design |
| Behavioural interview ("tell me about a time") | Storytelling and interview prep | Rewards rehearsed narrative over real skill |
| Paid work sample, scored | Actual competency under mild pressure | None structural — quality depends on rubric rigour |
| Unpaid "trial project" | Willingness to work for free | Selects for desperation, not skill; repels strong passive candidates |
The row that should worry you is the second one, not the first. Everyone already half-suspects reference checks are soft theatre. Fewer people have clocked that the STAR-format behavioural interview they've built their whole process around has the same defect wearing a better suit — it measures how well someone can narrate a discovery call, not whether they can run one. If you want a script for the live version rather than the story version, the Cold Call & Discovery Roleplay Interview Script is built for exactly that swap.
What to keep the reference call for
I'm not saying skip it entirely. Keep a short one, but retask it: not "is this person good," which it cannot answer, but "is there anything in this person's employment history I should know before I make an offer" — gaps, terminations, the kind of factual, checkable claim a reference can actually confirm or deny rather than editorialise on. A tight version of this exists in the Reference Check Script for Sales Hires, and it should take ten minutes, not the forty-five most leaders still spend on it.
The forty minutes you save get better spent watching someone actually try to book a meeting with a prospect who doesn't want one. That's the only rehearsal that looks anything like the job.