<aside> ⚖️
FairScreen is a hiring assessment layer that treats fairness auditing as a core product surface. Every screening decision is logged, every position gets a live four fifths rule dashboard, and the system pauses positions that drift toward adverse impact before a regulator or lawsuit finds it first.
</aside>
Status: Draft v1.0 | Author: Anwesha Gupta | June 2026
FairScreen is a resume and assessment screening product for high-volume hiring. Unlike incumbent vendors that optimize purely for recruiter efficiency, FairScreen makes legal defensibility and fairness monitoring first-class features. It recommends candidates against a skills rubric, measures adverse impact per position in real time, and automatically routes decisions to human review when statistical thresholds are crossed.
One sentence: HireVue with a conscience and a dashboard to prove it.
<aside> 🔬
Research basis: Bommasani, Bana, Creel, Jurafsky, Liang. "Algorithmic Monocultures in Hiring." FAccT 2026. The largest empirical study of deployed hiring AI: 3.4 million applicants, 4 million applications, 156 employers, one shared vendor.
</aside>
Over 90% of U.S. employers use algorithmic screening, and most buy from the same few vendors. Over 60% of the Fortune 100 use HireVue alone. This creates an algorithmic monoculture: one model's blind spots become every employer's blind spots.
| Finding | Implication |
|---|---|
| 25.87% of Black applications and 14.74% of Asian applications routed to positions showing adverse impact under Title VII | Bias exists in deployed systems at scale, today |
| Adverse impact invisible in aggregate, visible only position by position | Annual aggregate audits are theater; measurement must be per position |
| Systemic rejection rates far exceed the independent decisions baseline | Rejection at one employer predicts rejection everywhere using the same vendor |
| Applicants need ~25 applications (vs 10 under independent decisions) for 99.9% chance of one recommendation | Monoculture taxes applicants, especially adversely impacted groups |
<aside> 💡
Problem statement: No screening product on the market makes position-level adverse impact monitoring a real-time, decision-gating feature. Compliance teams discover disparities years late, through subpoenas instead of dashboards.
</aside>
| Persona | Role | Pain today | What FairScreen gives them |
|---|---|---|---|
| Priya, TA Lead | Runs 40 open requisitions, 800 applications per week | Needs algorithmic speed but fears legal exposure | Screening throughput with a defensibility layer |
| Marcus, HR Compliance Officer | Owns EEOC and NYC Local Law 144 obligations | Annual audits are manual, weeks of spreadsheet work, always backward looking | Live ratios per position, one click audit reports |
| Dana, Hiring Manager | Owns one Senior Data Analyst requisition | No idea if her shortlist would survive scrutiny | Position health status and a reviewed, defensible shortlist |
| Wei, Applicant (indirect user) | Applying to 25+ roles | Correlated rejections from shared vendor models | Statistical drift in her demographic pauses auto rejection and triggers human review |
Each use case is a toggle. Primary actor, trigger, flow, and acceptance criteria inside.