Building RecruitAI forced us to separate AI features that save recruiter hours from features that demo well and die in production. The dividing line is simple: does the AI remove a repetitive step, or does it try to replace judgement?
What works
Resume parsing into structured, searchable data. JD analysis that generates boolean search strings. Matching that shows its reasoning — matched skills, missing skills, experience gaps — so a recruiter can defend the shortlist to a client. Duplicate detection across vendor submissions.
What doesn't
Fully automated rejection. Personality inference from resumes. Auto-sent outreach with no human review. Anything where the AI makes a final decision about a person is both ethically shaky and commercially risky — clients want explainable shortlists, not black boxes.
Our rule: AI drafts, humans decide. Every score in RecruitAI is explainable, and every automated step has a review point.