Explain AI screening to a client by showing three things on every submission: match percentage, which must-have skills are present or missing, and recruiter notes from the confirmation call. Clients hate black boxes. A 62% with confirmed notice and CTC beats an unexplained 81%. The score is a sort key; the notes are the decision. That is the whole vendor-call story.
“The AI shortlisted them” is how you lose a sceptical hiring manager in one sentence.
Clients reject black boxes, not ranked lists
Indian hiring managers have been burned by keyword-stuffed CVs and by agencies who dump volume. If you now say “our AI ranked 325 people from 6,355 applications,” they hear a bigger dump. Ranking is how you cut 4–6 hours of triage. It is not how you earn trust. Trust is: we ranked, we called, we are sending these four, here is why each one survived. The funnel diagnostic behind a 45% shortlisting rate is in what a 45% client shortlisting rate tells you. Use it as a conversation, not as a boast.
How the percentage is built — parse, extract, gate, rank — is in how AI resume screening actually works. You do not recite the pipeline on a vendor call. You recite the outputs the client can check against their JD.
The three things to put on every submission
| Field | What the client hears | If you skip it |
|---|---|---|
| Match % | How close the resume text is to this JD | “You guessed” |
| Must-haves present / missing | Whether Java (or their deal-breaker) is actually there | They find the gap and stop trusting the pack |
| Recruiter notes | Notice, CTC, location, intent confirmed | They assume you forwarded an unread PDF |
If a must-have is missing, say the 55% cap. Do not bury a gated profile in the middle of an 80% list and hope they do not ask. If you still send that person, the note has to be the reason: “Java missing on paper; candidate self-rated 8/10 and walked through a production service on the call; your call if you want to see them.” That is an override you can defend. See can I override an AI candidate score.
A vendor-call script that survives a sceptical hiring manager
“We rank every CV against your JD. A recruiter then confirms notice, CTC, and intent. You only see people who passed both.”
“This one is 78%. Must-haves: Java and Spring present. Notice 30 days, CTC inside band, willing for Hyderabad. That is why they are first.”
“Two higher paper scores failed the call on CTC. One was capped at 55% — no Java. We did not want those in your inbox.”
“If product-company only is a hard gate, say so. We will tag it as must-have instead of discovering it as a bounce.”
325 AI-matched names is your problem, not theirs. Sending the raw rank re-creates the resume flood on the client side and trains them to ignore you.
Match 74%. Must-haves: React, TypeScript present; AWS good-to-have missing. Recruiter call 31 Aug: 30-day notice, 14 LPA expected vs 12–15 band, Bangalore local, actively serving notice. Not a black box — paper plus confirmation.
What to say when the client asks “why not this profile?”
They will forward a Naukri link they liked. Do not get defensive about the model. Score it live if you can: match %, must-haves, then “we had not called them yet” or “we called; 90-day notice.” If their favourite is a title variant your old Boolean missed — SSE versus Software Engineer — say so, and say you now rank on meaning plus gates. If their favourite is missing a must-have, say the cap. If you overrode the rank for a referral, say the referral. Mystery is what they are punishing. A sentence with a reason is what they are buying.
Explainability is also how you keep using AI without looking like you outsourced judgement. Rex-quality shortlisting at 64% versus a looser AI pile at 40% is often a confirmation gap, not a maths gap. Narrate confirmation, and the percentage becomes a feature instead of a fight.
Shortlists you can narrate on a client call
TeamOB - SafalHires surfaces match percentage, must-have gaps, and room for recruiter notes — so the vendor call is a walkthrough, not a shrug.
Free trial