Enterprise recruiting fails in the first mile. A strong JD goes live; applications flood the ATS; screeners skim for keywords under time pressure. Strong candidates with unconventional formats are skipped. Weak keyword matches advance. Hiring managers lose trust in the shortlist and restart the search informally through networks — widening inequity and slowing time-to-offer.
Manual screening also resists measurement. Nobody can explain why two similar candidates diverged after the CV stage. When diversity or labour regulators ask about process fairness, the answer is “we tried our best” backed by sparse notes. Agency CVs and internal referrals add another channel that never entered the same criteria.
Seasonal hiring and multi-country roles amplify the mess. Language variants, degree naming differences, and skill synonyms defeat brittle keyword filters. Recruiters become copy-paste engines scheduling interviews for whoever survived the chaos — not who best matched the role scorecard.
Talent acquisition owns the scorecard process; hiring managers own must-haves; legal and DEI partners own fairness constraints; IT owns ATS connectors. Anti-patterns include scoring “culture fit” without definition, training on who was hired last year as ground truth, and auto-rejecting without recruiter visibility.
AI screening should compress the first cut with consistent, role-specific criteria and transparent scoring — then hand a defensible shortlist to humans for interviews. It should never silently reject people from a black box with no recruiter override.