// ARTIFICIAL INTELLIGENCE
Voice-first hiring: ATS versus talent intelligence
7 min readnijitech
An ATS is a system of record; talent intelligence is a system of assessment. Treat them as the same thing and candidates who could do the job get cut on keyword match.
Full post
Four hundred applications arrive for one opening. The system narrows them to forty, and the recruiter talks to those forty. Among the three hundred and sixty who were cut are people who could do the job — they just wrote their CV in different words.
The problem is not that the ATS works badly. The problem is what an ATS is: a system of record. Ask a system of record to assess, and matching text starts to look like capability.
What an ATS does
An applicant tracking system receives the application, breaks the CV into fields, compares it against the words in the posting, places the candidate at a stage and keeps a record of all of it. These are real jobs, and doing them by hand is not possible.
Four things an ATS genuinely solves
- Applications collected in one place and not lost
- CVs turned into structured fields
- Visibility into which stage each candidate is at
- A record of who was rejected and why
None of those four answers the question that matters: can this person do this job? The ATS does not ask it, because it was never built to.
The limit of keyword matching
A CV is a self-report — a text the candidate wrote after reading the posting and guessing which words would be searched for. The match score measures the accuracy of that guess, not the candidate’s capability.
The consequence: two candidates who did the same work for the same length of time get different outcomes because one wrote “process improvement” and the other wrote “operational efficiency”. As CV-optimisation tools spread, both sides optimise the text and the signal gets weaker still.
Why a conversation carries a different signal
In a structured interview the questions are asked to everyone in the same order, the answer is produced live and cannot be edited. Where the CV is thin, a follow-up can be asked — the thing a document cannot do.
Four things a conversation carries that a CV does not
- How the candidate describes their own work — which decision, and why
- Whether they say when they do not know
- Whether the answer deepens when a follow-up arrives
- That the same question was put to everyone — interviewer variance removed
The red line on automated assessment
An automated system can rank, summarise and flag. It cannot reject on its own. The EU AI Act places AI systems used in employment and candidate selection in the high-risk class; human oversight, transparency to the candidate and record-keeping are the conditions of that class.
In practice that means three things: the candidate knows AI is part of the process, the reasoning behind every assessment is recorded, and a human makes the final call.
Bias is not only in the model
Debate about bias in AI hiring usually focuses on the model’s training data. But bias forms in three places at once: in historical hiring data, in the question set itself, and in the scoring criteria.
Fixing the question set does not solve all of that, but it solves one part: it removes the variation in questions from interviewer to interviewer. Without a common measure, comparison is not possible in the first place.
Summary
System of record versus system of assessment
- An ATS stores the application; talent intelligence assesses the candidate
- A match score measures word-guessing, not capability
- A conversation is produced live — it cannot be edited, it can be probed
- The system ranks; a human rejects, and writes down the reason
Products mentioned in this post
From the glossary: ATS (applicant tracking system) · Explainable AI (XAI) · Human-approved decision (human-in-the-loop)