Most poor AI screening results come from a thin setup, not a weak tool. Ten minutes of structured prep on the mandate, with a clean brief, ranked criteria and a small calibration batch, decides whether your first 200 CVs come back as a usable shortlist or a pile you re-read by hand anyway.

Picture Monday, 8:40am. A new mandate landed on Friday afternoon, 200 applications are already in, and the client wants a shortlist by Wednesday. You have two other live roles. If you skim a CV in 90 seconds, that's 5 hours before a single qualification call. So you upload everything, hit go, and get results that feel off. The cause is almost always upstream, in what you fed the system.

Where new mandates go wrong

The failure usually shows up in one of three places.

First, the brief you paste in is the client's job ad, which is written to attract candidates, not to screen them. It's padded with culture language, lists fifteen "requirements" as equals, and never says which three actually decide the placement.

Second, nobody separates must-haves from nice-to-haves. When everything counts equally, a candidate with seven of ten skills can outrank someone who has the two skills the hiring manager really cares about.

Third, nobody checks the output before the full batch runs. You find out the criteria were off after you've already made decisions on the results.

A checklist fixes all three. Run it at intake, before the CVs arrive if you can.

The setup checklist

1. Rewrite the brief for screening, not advertising. Start from your intake call notes, not the public ad. Strip the marketing copy. Keep role scope, seniority, sector context and the real reason the seat is open. If you want a head start, the job description templates give you a clean structure, and structuring your job brief so AI screening actually works covers the thinking in more depth.

2. Pick your three to five deal-breakers. Ask the hiring manager: "If a candidate lacked this, would you still interview them?" Anything they'd still interview without is not a must-have. Typical deal-breakers are a licence or certification, a specific tool at production level, or a minimum scope of team or budget managed.

3. List the nice-to-haves separately. These break ties between strong candidates. Keep them to five or fewer, or they'll drown out the must-haves.

4. Write skills as skills, not job titles. "Account Manager" means different things at different firms. "Manages renewals on multi-year enterprise contracts" is something a screen can actually match against. The guide to 50 skills to screen for is a useful prompt list when you're stuck describing what the role needs.

5. Set the experience range, then sanity-check it. Clients often say "5+ years" when they'd happily see someone with 3 who has done the harder version of the job. Ask them directly, and record the answer.

6. Note location, notice period and salary constraints separately. These are logistics. Keep them out of the skills criteria so a great candidate with a long notice period isn't scored down on ability.

7. Run a calibration batch of 10 to 15 CVs. Include a few you already know are strong, a few you know are weak, and a couple of borderline cases. Where the results contradict your own read, adjust the criteria and rerun. This one step does the most for accuracy.

8. Read the reasoning, not just the score. Check why a candidate ranked high or low. If the explanation cites the wrong evidence, your brief is ambiguous at that point. Fix the wording, not the candidate.

9. Run the full batch, then spot-check the edges. Open the top ten and also the ten just below your cut-off. If the near-misses look better than the bottom of your shortlist, your threshold is too tight.

10. Save what worked. Keep the final criteria as a template for similar mandates. Your second role in the same sector should take a fraction of the setup time.

The human layer

Screening tells you who matches the brief on paper. It can't tell you why someone left three roles in four years, whether they're talking to two other agencies, or whether their "led a team of eight" was a title or a reality. It also can't judge whether this person will survive a hiring manager who interrupts every answer.

That's what the 20-minute qualification call is for. Use the screen to decide who gets a call, not who gets the job. For a framework on where that line sits, read what makes a candidate "qualified" after an AI screen. And when a role is senior or highly relational, a manual search may still serve you better than any screen.

Your first step on Monday

Don't try all ten steps on your most critical mandate. Pick the next role that comes in, spend fifteen minutes on steps 1 to 3 at intake, and run a calibration batch before the full pile. Compare how long the shortlist takes against your usual process.

When you're ready to test it on a live role, you can Start Free Trial and put CV Matcher through this checklist with a real brief.