Explainable AI in HR is the requirement that an AI tool used for people decisions can show, for a specific output, which factors drove it, in terms a recruiter, manager or candidate can follow and check. A shortlist rank of 7 out of 300 means little until the system can say why. Without that, you cannot test a tool for algorithmic bias or defend its result.

What a usable explanation looks like

There are two levels. A global explanation describes how the model works overall: which inputs it uses and how much each weighs. A local explanation covers one decision: “ranked low because the role asks for SAP FICO and the CV shows two years, against a five-year requirement.” HR needs both, the first to approve the tool inside your applicant tracking system and the second to answer a candidate or a hiring manager.

NIST’s 2021 report on four principles of explainable AI is a good yardstick. The system should give reasons with its output; the reasons should make sense to the person receiving them; they should accurately reflect how the output was really produced; and the system should flag when it is outside the conditions it was built for. The third point catches many products: a neat explanation generated after the fact can describe something different from what the model actually did.

Why hiring tools face the hardest test

A rejected candidate, a manager who disagrees with a screening result, or an internal auditor can all ask “why?”. If the only answer is “the model said so”, HR cannot correct an error or show the decision was job-related. Abroad, rules point the same way. New York City’s Local Law 144 (enforced from July 2023) requires bias audits and candidate notice for automated hiring tools. The EU AI Act classes recruitment and worker-management AI as high-risk and gives affected people a right to an explanation of decisions made with such systems; after the 2026 Digital Omnibus amendment, those high-risk duties apply from 2 December 2027. Neither law binds an Indian employer hiring in India, and India has no AI-specific hiring law; MeitY’s November 2025 AI governance guidelines are voluntary.

Frequently asked questions

What should we ask a vendor?

Ask for a sample explanation for a real rejected profile, the list of inputs the model uses, and whether factors like college or location are excluded; pair the answers with your interview scorecard criteria to see if they match the job.

Are generative AI tools explainable?

Not in the same way; a generative AI assistant can cite its source document, which helps, but its reasoning about a candidate is not reliably traceable, so keep it away from final ranking.

Does explainability reduce accuracy?

Sometimes slightly, but for a hiring decision a simpler model you can defend usually beats a marginally sharper one you cannot.

Explanations matter most where a tool ranks people, as in AI video interviews and talent intelligence platforms.