What evidence should be kept?

Separate each decision stage so early screening bias is not hidden by later interview outcomes. Run group-level selection rates, impact ratios, confidence notes, and threshold sensitivity checks. Compare model, human-review, and final-outcome results to find where disparity enters the funnel.

algorithmic bias testing

Algorithmic Bias Testing for Hiring Tools

Algorithmic bias testing checks whether a model, rule set, or scoring workflow produces systematically different outcomes for groups that should be reviewed under employment law and internal fairness policy.

Run audit preview

Where this fits

A vendor introduces a new matching score and the employer wants to validate outcome impact.

A team changes pass thresholds for assessments and needs to see group-level effects.

An internal analytics group wants repeatable testing instead of one-off spreadsheets.

Operating steps

Separate each decision stage so early screening bias is not hidden by later interview outcomes.
Run group-level selection rates, impact ratios, confidence notes, and threshold sensitivity checks.
Compare model, human-review, and final-outcome results to find where disparity enters the funnel.
Review flagged cells with HR, legal, and the process owner before changing production thresholds.
Retest after remediation and preserve both the original result and the follow-up result.

Common risks

Testing only model accuracy while ignoring how errors distribute across groups.
Using too little data and overstating certainty in the report.
Changing thresholds without documenting why the change reduces risk without hurting job relevance.

How HireBias Audit connects

HireBias Audit packages algorithmic bias testing into an audit console with matrix views, 4/5 rule checks, and exportable evidence.

View plans