Screening Score Predictive Validity Index (SSPVI)

GMI-HCM-002Human Capital ManagementIntegration-RequiredPattern O-1: Confidence Calibration

Definition

The correlation between AI screening scores and actual job performance outcomes, computed at each score band to detect where predictive power breaks down.

The Measurement Gap

Every AI screening tool produces a score, a rank, or a pass/fail decision. Organizations use these outputs to filter thousands of candidates, trusting that higher scores predict better job performance. But almost no organization systematically validates that assumption. The screening score becomes an article of faith: it governs who enters the pipeline, but no metric checks whether the score actually correlates with what happens after hire.

Legacy metrics cannot detect this gap. Time-to-fill measures pipeline speed, not screening accuracy. Quality-of-hire metrics, where they exist, rarely connect back to the specific screening score that admitted the candidate. Offer acceptance rate and new-hire retention measure downstream events without isolating the screening signal. An organization can operate an AI screener with near-zero predictive validity (high scores and low scores performing identically on the job) and never see it on any standard HR dashboard. The wasted cost of false confidence in screening scores cascades through the entire talent acquisition function: interview capacity spent on mis-ranked candidates, hiring manager frustration with candidate quality, and systematic exclusion of candidates whose attributes the algorithm weights incorrectly.

Without band-level validation, even a screener with moderate overall correlation can hide pockets of complete predictive failure: score ranges where the algorithm is essentially random.

Decision Supported:

Do our AI screening scores actually predict job performance, and at which score ranges does predictive power break down?

Formula

Plain language: For each screening score band, calculate the correlation between the AI screening score and a composite job performance measure. Then compute the overall SSPVI as the weighted average across bands.

Notation:

SSPVI(b) = r(screening_score, performance_composite) for each band b

Overall SSPVI = Σ (n_b / N) × SSPVI(b)

Variables:

Variable Definition Unit
screening_score The numeric score or rank assigned by the AI screening tool at time of application Score
performance_composite A composite measure of job performance at 12 months post-hire, combining manager ratings, objective output metrics, and retention status Score
b Score band, a defined range of screening scores (e.g., deciles or quartiles) Category
n_b Number of hired candidates in band b with 12-month performance data Count
N Total number of hired candidates with 12-month performance data across all bands Count

Proposed formula, pending validation pilots.

At a Glance

Unit
Correlation (-1 to 1)
Trend Direction
Higher
Reporting Frequency
Annually (requires 12-month performance outcomes)
Category
AI Screening & Pipeline Integrity
← Screening False Negative Estimation Rate (SFNER)AI Assessment Anchoring Bias Index (AABI) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment