Why a Human Final Hiring Decision Is Not Enough

Automation overreliance happens when a hiring team treats a system’s output as sufficient evidence to advance or reject a candidate—without checking the underlying information, job relevance or limitations. It can involve AI rankings, ordinary rule-based filters or automated assessment cutoffs.

The important question is not simply “Does a human make the final hiring decision?” It is “Who checks the people the system removes before that decision?”

Consider a hypothetical workflow: a tool ranks 400 applications, a recruiter reviews only the top 40, and a manager chooses whom to interview. Humans make the visible decisions, but the ranking determines whose evidence they see. Reviewing the shortlist alone cannot reveal qualified applicants excluded below it.

1. A recommendation becomes an untested cutoff

A fit score may be intended to support comparison, not justify automatic rejection. Turning it into a cutoff changes how the tool is used—and demands evidence for that use.

In its November 2024 UK recruitment-tool audit report, the Information Commissioner’s Office (ICO) said the tools it reviewed were intended to support human recruiters, not make recruitment decisions without human intervention. It recommended preventing progression or rejection based solely on indicative grades or fit scores. That finding concerns the audited tools, not every hiring product. See the ICO audit report, pages 32–33.

What HR should check: What does the score measure? Which inputs produced it? Was it validated for this role and decision? Do the contract, configuration and recruiter training agree on its intended use?

In the United States, the EEOC says selection procedures should be validated for the positions and purposes for which they are used. Vendor documentation may help, but the employer remains responsible for ensuring validity under the Uniform Guidelines on Employee Selection Procedures. A vendor’s general accuracy claim is not a substitute for that check. See the EEOC’s testing guidance.

2. Human review becomes confirmation rather than evaluation

A reviewer who sees only a score and summary may confirm the recommendation without examining the candidate’s actual evidence. Adding an approval button does not supply missing information, time or authority.

The ICO’s 2024 audits found that where human reviews were not formalized, staff were typically untrained and reviews were not consistent or thorough. It recommended documented review methods, training and records of changes and reasons. See the ICO audit report, pages 32–33.

Make the review substantive:

• Give reviewers the original application or assessment response, not just the output.
• Require a job-related reason for advancement or rejection; “low AI score” is not an explanation of the candidate’s evidence.
• Give reviewers authority to pause the action, correct inputs and override the recommendation.
• Where practical, have reviewers assess the evidence before seeing the score, then investigate disagreements.

Human judgment also needs constraints. Use the same role criteria across candidates rather than replacing automated ranking with an unstructured “gut feel.”

3. The workflow mistakes an access barrier for poor performance

An assessment result can reflect difficulty using the assessment rather than inability to do the job. Automating the resulting rejection makes that distinction easy to miss.

The U.S. Department of Justice explains that hiring technologies can unlawfully screen out qualified people with disabilities. Its examples include facial or voice analysis disadvantaging applicants with autism or speech impairments, and an online interview incompatible with a blind applicant’s screen reader. These examples do not establish that every video-interview tool uses those features. See the DOJ’s disability and hiring-technology guidance.

What HR should change: Explain the assessment format before it starts and provide a clear accommodation contact. As an operational control, pause automated rejection while an accommodation request or reported technical failure is being resolved.

For covered U.S. employers, reasonable accommodation during hiring is an ADA obligation unless it would create undue hardship—not merely a candidate-experience feature. See the DOJ guidance. Build a clear route from automation to a person.

4. A faster funnel hides selection errors

Time saved shows efficiency, not whether the system identifies suitable candidates. If the team checks only people who advance, it has little evidence about qualified people rejected earlier.

Monitor the exclusions as well as the shortlist:

• Sample rejected applications, including some well below the cutoff, against the same job-related rubric.
• Separately review ambiguous inputs, missing scores and disputed information.
• Track errors, overrides, candidate withdrawals and selection-stage disparities where lawful and appropriate.
• Recheck after changes to the model, weights, cutoff or role requirements; retain the version and decision records under an appropriate retention policy.

Random and risk-based review, including sampling after algorithm changes, are approaches recommended in the ICO’s 2024 report. Samples can expose problems; they do not prove that every rejection is correct or satisfy every legal duty. See the ICO audit report, pages 32–33. Use a vendor validation checklist before deployment, not only after complaints.

What candidates can ask

Ask whether a score determined advancement, whether application information can be corrected, and who handles assessment-access problems. Describe a specific error or barrier and request review against the published job requirements. Keep the relevant notice, submitted application and correspondence.

These are practical requests, not a universal entitlement to a score or human reconsideration. A rejection alone also does not prove automation caused it.

Reviewed October 27, 2026. Informational only—not legal, HR or employment advice. The U.S. obligations and historical UK audit findings above are distinct; confirm applicable requirements with qualified counsel.