When you hire, you're not describing a candidate. You're betting on the future: will this person succeed in this role, six months from now, a year from now? Hiring is a prediction problem, and that framing changes everything.
The problem is that most processes treat hiring as a description problem: check a CV, confirm a good impression. None of those steps measures what actually matters, future performance. This article reframes hiring as a prediction, shows what research says about how reliable each method is, and explains how to build a process that predicts instead of describes.
Table of contents
1. Hiring Means Predicting Future PerformanceA hiring decision is a prediction in disguise. You observe signals in the present (a CV, an interview, a test) to bet on a future outcome you won't see until the person is on the job.
Framed this way, hiring follows the same rules as any prediction:
The problem is that most recruiters have never framed their job this way. They optimize the confidence felt during the interview, not the accuracy of the prediction.
Predictive validity is a method's ability to forecast a candidate's future performance. It's measured as a correlation between the assessment score and the performance observed 6 to 12 months after the hire.
In practice, to know whether your process predicts:
What matters is that correlation, not the evaluator's feeling. A method can inspire confidence and predict badly. That's exactly the case for most of the most widely used tools.
The methods that dominate hiring are often the least predictive. The reference meta-analysis by Schmidt places the validity of unstructured interviews and CV reviews in a low range, around r = 0.10 to 0.38.
Several reasons for this:
The result? If you rely mostly on CVs and classic interviews, your process predicts weakly, no matter how confident your managers feel. Felt confidence is not a sign of validity.
The same research shows what predicts well. The strongest single predictors are real-world tests and structured interviews:
And crucially, combining relevant methods predicts better than any one alone: pairing a cognitive test with a structured interview pushes combined validity to around R = 0.63.
What matters is the principle: you predict performance better by watching a candidate act in conditions close to the role than by reading what they claim to know. Platforms like Scalyz are built precisely on this logic of standardized real-world scenarios.
Treating hiring as an explicit prediction problem doesn't require a statistical model. Three principles are enough.
Replace what reassures (the CV, the personality chat) with what predicts (a work sample, a structured test). Evaluator comfort is not a criterion.
A prediction only means something if every candidate is measured on the same scale. A shared scoring rubric and an identical scenario for everyone are the baseline condition.
This is the step almost nobody takes: comparing assessment scores to real performance at 6 months. That loop is what turns a frozen process into a system that learns. The IT hiring decision tree formalizes this traceability.
An IT services company (ESN) hires consultants on CVs and interviews. Over a year, a third of its placements get challenged by clients. The usual reflex: make the interviews tougher.
The predictive reflex: treat the process as a model to validate. The team collects assessment scores and compares them to manager feedback at 6 months.
The result:
The correlation is near zero, the interviews predicted nothing. After switching to standardized real-world scenarios with a scoring rubric, the same measurement shows a strong correlation between score and performance. The process became predictive, and the rate of challenged placements collapses.
It's a selection method's ability to forecast a candidate's future performance, measured by the correlation between the assessment score and performance observed several months after the hire.
Work sample tests and structured interviews are among the strongest single predictors. Combining a structured test with a structured interview predicts even better than either alone.
No. A CV mostly measures past experience, a weak predictor of future performance according to research. Use it to screen, not to decide.
By closing the measurement loop:
As long as hiring is treated as a description of the present, it stays driven by intuition and bias. Reframed as a prediction problem, it becomes measurable, comparable, and improvable.
The question is no longer "does this candidate make a good impression?" but "does what I measure actually predict their success?" That's the only question that separates reliable hiring from an expensive bet.
Want to make your hiring genuinely predictive? Book a Scalyz demo.
Share this article :