Validity is an essential component of any candidate assessment or evaluation measure. AI or not, the standards haven’t changed — but in the absence of the transparency and explainability of most AI hiring systems, finding the truth has become a challenge.
Learn the facts about validation.
Download our Validation Briefing →Validity is the degree to which a test measures what it’s supposed to measure — and predicts what it claims to predict. Validation is the research process that produces credible evidence for both. Without it, hiring decisions rest on irrelevant information — you might as well flip a coin.
Strip away the complexity and it comes down to two things: job-relatedness and accuracy — each can be verified through two core methods.
Documenting the human attributes required for success — most commonly through a research-based process called job analysis.
Proving a tool measures only job-related attributes with a direct, measurable impact on performance — typically through a statistical study correlating scores with performance outcomes.
An assessment that has not been validated is a guess with a score attached. It can look precise and produce clean numbers while measuring something that has little to do with the job. The ratings it hands you may simply be wrong — and because people trust a number, you hire the wrong person.
Job-relatedness and legal defensibility are two sides of the same coin. Why would you use information that is not related to the job to decide who gets hired? Under the Uniform Guidelines, a selection procedure that produces adverse impact has to be shown to be job-related and validated. Without that evidence, there is nothing to show.
Hiring on a tool that does not predict performance is expensive in the places that are hardest to see — turnover, time to proficiency, and the strong candidates you screened out without knowing it.
When a decision is challenged — by a candidate, a regulator, or a court — the question is whether the tool is both fair and accurate. Validation is the evidence that answers it.
Plenty of vendors claim their measures are “validated.” Most fail to prove it.
What to demand when evaluating vendors