Audit a hiring process for fairness

---
name: audit-a-hiring-process-for-fairness
category: get-hired
description: Audit an end-to-end hiring process for job relevance, consistent treatment, accessibility, subgroup outcomes, data quality, and effective remediation. Use when an organization must evaluate whether recruiting and selection practices create unjustified barriers.
---

# audit-a-hiring-process-for-fairness

Audit the decisions people experience, not only a final diversity percentage.

## When to use

- Use for role families, recruiting channels, screening, interviews, tests, references, offers, or vendor-supported hiring across one or more jurisdictions.
- Use `audit-an-automated-hiring-system` when the primary scope is one automated decision system; include it here as a dependency when automation affects the wider process.

## Preconditions

- Establish independent audit authority, qualified employment and privacy guidance, worker and disability perspectives, security controls, purpose, jurisdictions, time period, and remediation ownership.
- Preserve requisitions, job analyses, criteria, rubrics, applications, decisions, overrides, accommodations, recruiter notes, interview records, test validation, vendor documentation, complaints, offers, and outcome data.
- Define lawful data access and minimum subgroup reporting thresholds. Do not infer protected traits for individual decision-making or expose small groups.

## Procedure

1. Build a **decision and data inventory** from workforce planning through sourcing, application, screening, assessment, interview, reference, offer, negotiation, withdrawal, and onboarding.
2. For every stage, record owner, inputs, rule, discretion, automated support, data source, retention, accommodation, output, appeal, and downstream consequence.
3. Trace each criterion to current **job-related criteria and validity** evidence: work analysis, required outcome, measurement reliability, predictive or content validity, alternatives, and known limitations.
4. Test whether stated requirements are actually necessary at entry or can be learned with reasonable support.
5. Reconstruct selection outcomes by stage, role, location, channel, recruiter, interviewer, assessment, accommodation, and other legitimate operational segments.
6. Perform **subgroup and outcome analysis** using traceable candidate-to-application-to-stage denominator lineage, confidence or uncertainty, intersectional views where safe, missingness, multiple comparisons, and practical harm alongside statistical measures.
7. Inspect differential drop-off, scoring, overrides, time-to-response, accommodations, scheduling, negotiation, offers, and withdrawals rather than analyzing hires alone.
8. Sample decision files and observe practice to find inconsistent rubric use, inaccessible steps, proxy criteria, stereotype-based notes, conflicts, and unauthorized data.
9. Test vendor, referral, advertising, background, and automated-tool boundaries, including who can explain, correct, appeal, and delete data.
10. Investigate candidate complaints and compare lived barriers with aggregate results. Protect complainants and do not expose them to hiring decision-makers unnecessarily.
11. Check aggregation reversal across country, source, stage, role variant, accommodation, and hiring cohort before reporting any overall rate. Prioritize causes by harm, affected stage, evidence, and controllability.
12. Implement **remediation and monitoring**: remove unjustified criteria, repair access, protect deadlines, standardize evidence, retrain decision-makers, pause unsafe tools, rerun affected decisions where possible, and assign candidate remedy.
13. Backtest changes, monitor stage-level outcomes and overrides, audit exceptions, and set explicit reopening triggers.

## Failure plan

- If data cannot support a reliable subgroup conclusion, report the gap and strengthen collection or qualitative evidence instead of declaring fairness.
- If an active step lacks job relevance or blocks accommodation, pause it where authorized while an equivalent path is established.
- If audit access exposes sensitive candidate data, isolate the dataset, reduce fields, and preserve an access trail.
- If leaders pressure the audit toward a predetermined conclusion, document the scope conflict and escalate to the independent sponsor.

## Worked example

A company hires one role family in eight countries using employee referrals, resume screening, a timed test, panel interviews, and a vendor ranking. Overall hiring rates look balanced, but disabled candidates abandon the timed test, one location applies an unwritten degree rule, overrides favor referrals, and demographic data is missing unevenly. The audit maps every decision, validates criteria against actual work, analyzes stage-level outcomes with uncertainty, reviews complaints and accommodations, removes unsupported barriers, retests affected candidates where possible, and monitors overrides and candidate harm after remediation.

## Done

- A verified hiring process and decision register records stages, owners, data, criteria, discretion, automation, accommodations, appeals, jurisdictions, and candidate consequences
- A validity and fairness analysis verifies job relevance, reliability, data quality, stage and subgroup outcomes, overrides, barriers, complaints, uncertainty, and alternatives
- A remediation and monitoring report records candidate remedy, control changes, validation, owners, deadlines, residual risk, exception review, and reopening triggers