Q: What are the first steps to detect bias in automated hiring tools? A: If your team asks how to detect AI hiring bias, start by measuring selection rates by race and gender at every stage and running matched-resume audits that vary only names, schools, or zip codes to compare pass rates. Inspect model features for proxies like zip code or school, set fairness thresholds, audit vendors, monitor results each month, and document actions. Q: Which metrics should teams measure first to spot unfair screening? A: Measure the selection rate ratio using the 80% rule by comparing pass rates for each group at every stage and flagging groups whose rate falls below 80% of another’s. Also check error rates by group (false negatives and false positives), calibration of scores across groups, drop-off points, and intersectional gaps such as combined race and gender groups. Q: What quick tests can my team run this week to check for bias in hiring tools? A: Run matched-resume audits with paired resumes that differ only by names, schools or zip codes and compare scores and pass rates, and perform counterfactual tests that hide or swap likely proxies to see score changes. Use feature-sensitivity tools such as feature importance or SHAP, score candidates with a simple shadow rules model for comparison, and review candidate experience for devices, time limits and accessibility. Q: How should demographic data be collected and protected during audits? A: Collect voluntary self-identification data with a clear stated purpose and secure storage rules, and explain that it will not affect hiring decisions. Protect privacy by restricting access, setting minimum sample sizes, reporting in aggregates, and auditing category consistency across systems. Q: What product or process changes can reduce unfair screening? A: Blind early screening by hiding names, photos and schools and adopt skills-first structured rubrics tied to key tasks to focus on relevant ability. Balance or reweight training data to improve representation, remove or cap proxies like zip code and school tier, apply fairness constraints or threshold tuning, and keep human reviewers to check edge cases and override when needed. Q: What vendor standards should be in procurement to enable bias checks? A: Require independent bias audits and shareable summaries such as model cards and data sheets, plus feature transparency, configurable weights and an API for testing. Put audit rights into contracts to allow matched-resume and counterfactual tests, and require logging of all decisions with reasons and scores for traceability. Q: How can teams set up monitoring and guardrails to catch bias over time? A: Build dashboards that track selection ratios, error rates and calibration by stage and group, and set alert thresholds and auto-pause rules such as a selection ratio under 0.8. Re-run audits monthly or per 1,000 candidates, collect candidate complaints and recruiter notes, and document incidents and fixes in postmortems. Q: What short-term (30-60-90 day) plan helps detect and fix hiring bias? A: Follow the 30-60-90 plan: days 0–30 map and measure by inventorying screening tools, enabling voluntary self-ID, running selection rate ratios and a small matched-resume test, and freezing risky features if gaps are large. Days 31–60 focus on fixes and validation with skills-first rubrics, blind early screens and retraining or retuning models with fairness constraints, and days 61–90 launch dashboards, publish a bias risk policy and RACI, schedule independent audits, and train recruiters and hiring managers.