Definition
The estimated percentage of AI-rejected candidates who would have met the hiring bar if evaluated by a qualified human assessor.
The Measurement Gap
When an AI screening tool rejects a candidate, that rejection is functionally invisible. The candidate disappears from the pipeline, and no downstream metric ever evaluates whether the rejection was correct. Time-to-fill measures how fast the pipeline moves but cannot detect whether speed came from screening out qualified people. Quality-of-hire metrics only evaluate candidates who were hired: they have zero visibility into the candidates who were never given a chance. Recruiter workload metrics measure volume, not accuracy. The result is a systematic blind spot: an organization can reject thousands of qualified candidates per year through algorithmic false negatives and never see it on any dashboard. The cost compounds invisibly (missed talent, reduced diversity, and pipeline homogenization) while every legacy metric reports green.
Legacy metrics like applicant-to-interview ratio and screening pass-through rate measure volume proportions but cannot distinguish between a tight screen that correctly filters and a miscalibrated screen that discards viable talent. Without a structured sampling protocol that evaluates a subset of AI-rejected candidates against the actual hiring bar, the false negative rate of any screening algorithm is unmeasured and unmeasurable.
Decision Supported:
Is our AI screening tool rejecting candidates who would have been successful hires, and at what rate?
Formula
Plain language: Sample a random subset of AI-rejected candidates, have qualified human assessors evaluate them against the actual hiring criteria, and calculate the percentage that would have passed.
Notation:
SFNER = (Qualified_in_sample / Total_sampled) × 100
Variables:
| Variable | Definition | Unit |
|---|---|---|
| Qualified_in_sample | Number of AI-rejected candidates in the sample who meet the hiring bar when evaluated by a qualified human assessor | Count |
| Total_sampled | Total number of AI-rejected candidates randomly selected for human re-evaluation | Count |
Proposed formula, pending validation pilots.