PAR-2026-0203-APEX
Apex Talent Solutions
ResumeScreen AI v3.2 — Bias Audit Report
Executive Summary
8
Categories meeting four-fifths threshold
3
Categories requiring monitoring (0.80-0.89)
2
Categories below threshold (<0.80)
This audit analyzed 12,847 applicants processed by ResumeScreen AI v3.2 between January 1, 2025 and December 31, 2025. The AEDT is a resume screening tool that assigns scores from 0-100 to job applicants based on qualifications match.
Overall, the tool demonstrates compliance with the four-fifths rule for most demographic categories. However, two intersectional categories (Black or African American females in Engineering roles and Hispanic or Latino males in Sales roles) show impact ratios below 0.80 and require remediation attention.
Impact Ratio Analysis — Race/Ethnicity
| Category | Applicants | Selection Rate | Impact Ratio | Status |
|---|---|---|---|---|
| White (Reference Group) | 5,234 | 28.4% | 1.00 | PASS |
| Asian | 2,891 | 29.1% | 1.02 | PASS |
| Hispanic or Latino | 1,892 | 24.8% | 0.87 | MONITOR |
| Black or African American | 1,654 | 21.2% | 0.75 | FLAG |
| Two or More Races | 847 | 26.1% | 0.92 | PASS |
| Unknown/Not Disclosed | 329 | — | — | Excluded |
Impact ratios calculated using the selection rate of the highest-performing group (Asian, 29.1%) as the reference. Statistical significance tested via Fisher's exact test (p < 0.05).
Remediation Recommendations
Immediate Action (0-30 days)
Audit keyword matching criteria for Black or African American applicants
Review the resume parsing algorithm for potential bias in how educational institutions and professional terminology are weighted. HBCUs and minority-serving institutions may be underrepresented in training data.
Implement human review for borderline scores
For applicants scoring 45-55 (borderline pass/fail), require human review before automated rejection. This adds a safeguard while root cause analysis is underway.
Short-Term Action (30-90 days)
Request vendor model retraining with expanded dataset
Work with ResumeScreen AI vendor to retrain the model with a more representative dataset that includes greater diversity in educational backgrounds and career paths.
Conduct stratified analysis by department
The aggregate FLAG status may mask department-level variation. Analyze Engineering, Sales, and Operations separately to identify if the disparity is concentrated in specific roles.
Long-Term Action (90-180 days)
Establish continuous monitoring pipeline
Implement quarterly bias checks with automated alerts when impact ratios approach the 0.80 threshold. This enables proactive remediation before annual audit deadlines.
Evaluate alternative AEDT vendors
If vendor remediation efforts do not yield improvement by next audit cycle, consider evaluating alternative resume screening solutions with documented fairness testing.
Get your audit.
Every Paritas audit includes the same depth of analysis, three-tier risk classification, and prioritized remediation recommendations you see here.
This report is permanently hosted by Paritas and cannot be altered or deleted by the audited organization.
Report content licensed under CC BY-NC 4.0.
paritas.ai | audits@paritas.ai