If an AI hiring tool appears to have screened out or downgraded a qualified candidate, trace the decision before changing the system: identify the hiring stage, model version, inputs, job criteria and cutoff that produced it. Then compare the tool’s record with the application, check outcomes by group and stage, and investigate accessibility and accommodation. A disparity metric is a reason to look closer—not proof that a tool is fair, accurate or legally compliant.
First, identify what went wrong
“Inaccurate” and “biased” describe different problems, although they can overlap. An inaccurate result may come from a parsing error, stale information, a missing field, a misconfigured threshold or a criterion that does not measure the job requirement it is supposed to represent. A biased result may involve different outcomes for groups of candidates, an assessment that creates disability-related barriers, or a criterion with a discriminatory effect.
As an Amazon Associate I earn from qualifying purchases.
A score, rank, recommendation or classification is an output to examine—not a self-validating judgment about a person. Start by recording the role, decision date, hiring stage, tool and model version, and the exact output and rule that affected the candidate. Note whether the tool screened, scored, ranked, classified or recommended candidates; these are distinct functions and can produce different kinds of error.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →If a live decision may be wrong, a prudent operational safeguard is to route affected applications for consistent human review while investigating, where feasible. That is a risk-control recommendation, not a rule expressly prescribed by the legal sources discussed below. Human review is not an automatic cure: reviewers should use the same job-related criteria and their overrides should be monitored.
#1 Best Overall
Follow a troubleshooting sequence
- Freeze the relevant decision context. Record the job, stage, date, tool and model version, configuration, threshold or ranking rule, and output at issue. Keep the version and settings stable while comparing outcomes if practical; otherwise, a system change can obscure what caused the difference.
- Reconstruct the candidate’s inputs. Check the fields the tool received, where they came from, how old they are, how they were transformed, how missing values were handled, and what qualifications the tool claims to assess. Compare its record against the actual application. A parsing failure, stale profile or absent value should not be mistaken for a lack of qualification.
- Trace the criteria to the work. For each scored feature and cutoff, find the written job criterion it is meant to represent. Ask whether the feature measures a capability actually needed in the role, whether a proxy may be standing in for a protected trait, and whether the criterion is applied consistently. EEOC guidance says selection criteria with a significant discriminatory effect must be job-related and consistent with business necessity; its national-origin guidance identifies objective written criteria that are communicated and applied consistently as a promising practice.
- Compare outcomes at each decision stage. For each stage, count who applied, who advanced, and who scored above the relevant threshold. Examine legally and contextually relevant categories, including intersections where data permits. Record category definitions, sample sizes, unknown or missing demographic counts, comparison population and thresholds so that any difference can be interpreted.
- Find the mechanism behind a gap. If outcomes differ, test whether the difference follows a particular input, criterion, cutoff, data source, role or assessment step. Where feasible, test whether correcting the suspected input or criterion changes the result for the applications affected. Before resuming or expanding use, retest both the suspected failure and overall job-related performance. These are recommended diagnostic steps, not technical procedures specified by NYC law.
- Check disability access and accommodations. Consider whether the assessment could screen out someone who can perform the job with or without reasonable accommodation. Check whether candidates have a way to request accommodation or an alternative process, and whether the tool asks for disability or medical information in a way that raises legal concerns. The EEOC and Department of Justice’s May 12, 2022 technical assistance warns about these risks.
- Document the investigation and monitor changes. Keep the question investigated, data and model versions, metric definitions, findings, decisions, overrides and remediation. Repeat relevant checks after a change to a criterion, cutoff, data pipeline or model version.
Measure group outcomes without overreading a metric
Outcome comparisons can reveal a problem, but they do not explain its cause or establish whether the tool predicts job performance. Review the data behind any result: sample size, missing and unknown demographic information, job family, decision stage, threshold and comparison population. A small or unstable group comparison should not be presented as conclusive, and passing one metric does not establish that a system has no bias.
| Measure | What it captures | How to interpret it |
|---|---|---|
| Selection rate | The share of relevant applicants or promotion candidates moved forward or assigned a classification. | Compare the same decision stage and relevant candidate pool; a rate alone does not explain why outcomes differ. |
| Impact ratio | For selection, a category’s selection rate divided by the most-selected category’s rate. For scores, a category’s scoring rate divided by the highest-scoring category’s rate. | It is one outcome lens, not a complete test of accuracy, job relevance, accessibility or compliance. |
| Scoring rate | The rate of people in a category whose score is above the sample median, under NYC Rules § 5-300. | Read it alongside the score distribution and the stage and job criteria involved; do not treat it as a direct measure of job performance. |
For covered automated employment decision tools (AEDTs), NYC Rules § 5-301 specifies bias-audit calculations that include selection rates and impact ratios for sex, race/ethnicity and intersectional categories. For a tool that classifies candidates into groups, the rule applies the calculations to each group as specified there; for scoring tools, it specifies the sample’s median score, category scoring rates and impact ratios. The rule also requires reporting the number of assessed people in unknown categories. An independent auditor may exclude a category comprising less than 2% of the audit data from required impact-ratio calculations, but the exclusion’s justification, applicant count and rate must be disclosed.
These NYC calculations apply in the scope set by the city’s rules; they are not a universal audit recipe for every employer or jurisdiction. The NYC rule’s definition of an AEDT’s simplified outputs includes scores, tags or categories, recommendations and rankings, while excluding tools that only translate or transcribe existing text. Confirm how the current rule applies to the tool and use in question.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWhen NYC Local Law 144 may apply
New York City Administrative Code § 20-871 restricts covered employer or employment-agency use of an AEDT to screen a candidate or employee for an employment decision unless the tool had a bias audit no more than one year before use and a summary of the most recent audit—including the distribution date of the tool to which it applies—was publicly posted before use.
Rank #3
The code also requires notice to covered NYC-resident candidates at least 10 business days before use. The notice must say that an AEDT will be used and identify the job qualifications and characteristics it will assess. Candidates may request an alternative selection process or accommodation. Information about the type and source of data collected and the data-retention policy must be available on the employer’s website or, if not posted there, provided on written request within 30 days, subject to legal exceptions.
The NYC Department of Consumer and Worker Protection’s FAQ, dated June 29, 2023, describes the law as applying where a job is located at an NYC office at least part-time, a fully remote job is associated with an NYC office, or the employment agency using the AEDT is located in NYC. It says covered use to substantially help assess or screen applicants at any point in hiring or promotion is included, while scanning a résumé bank or contacting someone who has not applied for a specific position is outside the described requirement. Because that FAQ is dated 2023, check for later city guidance before relying on its examples.
The same FAQ says Local Law 144 requires an audit but does not itself require a specific action based on audit results. That does not remove obligations under other anti-discrimination laws. An audit is not proof that a system is accurate, valid for a particular job, accessible or legally safe. Check the current city code and rules and get qualified employment-law advice for a specific situation; the legal sources here are not a legal opinion.
Apply employment safeguards beyond the audit
EEOC and DOJ technical assistance warns that employment software can screen out people with disabilities who could do the job with or without reasonable accommodation, and that AI assessments can raise concerns if they elicit disability or medical information. Review the assessment’s effects across different disabilities and ensure there is a workable accommodation process.
EEOC guidance on Title VII says criteria with significant discriminatory effects must be job-related and consistent with business necessity. Its national-origin guidance describes objective written evaluation criteria, communicated to candidates and consistently applied, as a promising practice. The applicable law and analysis depend on the protected trait, employer, role and facts, so consult current, topic-specific guidance rather than assuming one rule resolves every case.
When comparing more than one tool or process, assess them on the same dimensions: decision stage and output; evidence that criteria are job-related; outcomes by group and intersection; data source, completeness and age; accessibility and accommodation; human-review and override records; monitoring after changes; and any jurisdiction-specific audit and notice duties. Do not infer that one tool is safer merely because its score or audit summary looks different.
Sources and limits
The legal descriptions above refer to New York City Administrative Code § 20-871; NYC Rules §§ 5-300 and 5-301; and the NYC DCWP AEDT FAQ dated June 29, 2023. The disability discussion refers to EEOC and DOJ technical assistance announced May 12, 2022, and the EEOC’s national-origin guidance dated November 18, 2016. Verify current official text and agency guidance before acting, because legal requirements and guidance can change.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →These sources establish legal standards and procedural requirements, not a reliable general prevalence figure for how often AI candidate triage is inaccurate or biased. A particular audit or disparity result should be understood in light of its own tool, sample, criteria and context.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
