Playbook · 22 min · PDF, 18 pages
CV screening that actually ranks
Why most AI resume screeners fail silently — and the six evals every HR-staffing team should run before trusting one.
We audited 11 commercial and in-house CV screeners across 3 Talpro Universe products. Nine had silent degradation within 90 days. This playbook explains the failure modes, the six evals that catch them, and the operator-level numbers a vendor should be willing to produce.
What's inside
- 01Why ATS vendors benchmark wrong (and what to ask for instead)
- 02The six evals: rank correlation, recruiter-agreement, role-bias probe, timezone-drift, volume-spike, and cost-per-shortlist
- 03Golden dataset hygiene: how to build one from your last 500 hires without leaking PII
- 04The red-flag checklist: 14 things a vendor answer should NOT sound like
- 05Vendor vs. build vs. hybrid — the decision matrix we use on our own stack
“The most dangerous CV screener is the one that shows you a confident ranking on the day you buy it, then quietly drifts 9% per quarter because nobody wired up a drift eval.”
Written for
Heads of TA, recruitment ops leaders, HR-staffing founders evaluating their third AI screening vendor in a year.
Send me the PDF
One email. One PDF. No newsletter auto-subscribe, no drip. You can opt in to the monthly field-notes digest separately at the end.