Pre-employment assessment: what actually predicts performance

A pre-employment assessment is any structured test a company runs before an offer to predict how well someone will do the job: aptitude tests, personality questionnaires, skills tests, work samples, or a structured interview. Here's the position this guide takes. Most of them measure a proxy for the work. The ones that predict best measure the job directly, and the current evidence puts the structured interview first.

That distinction is the whole game, and most buyer's guides skip it. They hand you a catalog of test types and leave you to guess which one predicts anything. This guide leads with the evidence, then says where each type earns its place and where it just adds cost.

Key Takeaways

Most pre-employment assessments measure a proxy for the job. The best predictors are job-specific: structured interviews, job knowledge tests, and work samples.
Predictive validity is the only score that matters. In Sackett and colleagues' 2022 re-analysis, structured interviews rank first at about .42, work samples sit at about .33, and generic personality tests sit near the bottom.
Personality and aptitude tests are not useless, but sold as the default they are the weakest common choice, and cognitive tests carry the heaviest adverse-impact risk.
Fairness is not a separate box. A test that predicts performance and produces a defensible record is the one that survives an EEOC challenge, a NYC Local Law 144 audit, or an EU AI Act review.
When you buy a tool, check for reasoning per criterion, an audit trail, ATS write-back, and a human in the decision, not a bare pass or fail.

What a pre-employment assessment actually is

A pre-employment assessment is a standardized way to measure a candidate before you hire them, run the same way for everyone so the results compare. When every applicant faces the same task and the same scoring, you rank people on one scale instead of on how well each interviewer happened to click with them.

The labels move around. You will see the practice called pre-employment testing, and a single instrument called an employment assessment test. The vocabulary varies by vendor; the question underneath it does not.

The trap is treating "standardized" and "predictive" as the same thing. They are not. A test can be perfectly consistent and still measure something with almost nothing to do with the job, and a rigid one screens out people who could have done it well. Harvard Business School's Hidden Workers report estimated that automated screening pushes millions of qualified candidates out of the running.

So the real question is narrower than "is it consistent." It is: does the score predict who will actually do the job well?

The main types of pre-employment assessment, and what each measures

The category splits into a handful of families. Here is what each one measures, plainly, before we rank them.

Cognitive and aptitude tests

These measure general reasoning, numerical or verbal ability, and problem-solving speed. They are quick to run and predict performance moderately well on their own. The catch is that they tend to produce the largest score gaps across demographic groups, which is where adverse-impact risk concentrates. A high score tells you someone can reason, not that they can do this job.

Personality questionnaires

These score traits like conscientiousness and emotional stability, usually through self-report. They are popular because they are cheap to bulk-administer. On their own they are weak predictors, and self-report is gameable: candidates who guess what "good" looks like answer accordingly. Our fuller take on where these fit lives in the guide to psychometric testing for hiring.

Skills and work-sample tests

These have the candidate do a piece of the actual work: write and run code, debug a broken function, build a model, draft the email. Because the test is a slice of the job, the score maps directly to the thing you are hiring for. This family sits near the top of the evidence, just behind structured interviews and job knowledge tests.

Integrity tests

These predict counterproductive behavior such as theft and rule-breaking. They are more useful than their reputation suggests and pair well with a skills measure, but they answer a narrow question and are not a general fit test.

Structured interviews

A structured interview asks every candidate the same job-relevant questions and scores answers against a fixed rubric. Structure is the active ingredient, and it makes this the top-ranked method in the current research. The same conversation run loosely predicts less than half as well. We break down the mechanics in the structured interview software guide.

The only score that matters: does it predict performance

Every assessment claims to work. The way you check is predictive validity, a number between 0 and 1 that says how strongly the test score tracks later job performance. Higher is better. It comes from decades of research correlating pre-hire scores with real outcomes.

For decades the reference point was Schmidt and Hunter's 1998 meta-analysis, which put work samples on top at about .54. In 2022, Sackett, Zhang, Berry and Lievens revisited those estimates and found that earlier corrections for range restriction had inflated them. Most mean validities fell by .10 to .20, and the order changed. Here is the current ranking, with the older 1998 figure in parentheses:

Structured interviews: about .42 (was .51)
Job knowledge tests: about .40 (was .48)
Empirically keyed biodata: about .38 (was .35)
Work sample tests: about .33 (was .54)
Assessment centers: about .33 after the authors' 2023 update (was .37)
General mental ability (cognitive) tests: about .31 (was .51)
Integrity tests: about .31 (was .41)
Conscientiousness (personality): about .19 (was .31)
Unstructured interviews: about .19 (was .38)
Years of job experience: about .07 (was .18)

Reference checks and years of education dropped off the list, because the authors found too little usable data to re-estimate them. Read the top and the bottom together. The four strongest predictors are all job-specific measures, a point the authors make themselves, while generic personality self-reports sit level with an unstructured chat.

Cognitive tests fell from .51 to .31, and they carry the adverse-impact problem below. Work samples fell the most of any method near the top, but they still beat every generic proxy on the list. That is the evidence the personality-and-aptitude incumbents tend to gloss over, and plenty of work-sample vendors still quote the 1998 numbers too. If you are going to run a test, run one that measures the job.

Where personality and aptitude tests help, and where they mislead

None of this makes personality and aptitude tests worthless. Used well, they add signal. Conscientiousness has real predictive value as one input among several, and a cognitive measure can be a fair tiebreaker for high-reasoning roles. The mistake is making them the whole assessment.

Three failure modes recur. Self-report is coachable, so a personality score often measures how well a candidate reads the test, not who they are. A generic aptitude battery is role-blind: the same reasoning test gets sold for a sales rep and a backend engineer. And a cognitive score with a big group gap becomes legal exposure the moment it drives a rejection you cannot defend on job-relatedness.

The honest read: use personality and aptitude as supporting inputs if you like, but do not let them stand in for watching someone do the work. A proxy is a fine tiebreaker. It is a poor foundation.

Why structured interviews and work samples predict best

Structured interviews and work samples win for the same boring, powerful reason. They stay close to the job. A structured interview asks every candidate the same job-relevant questions against a fixed rubric, and a work sample has them do a small version of the work, so you are not inferring performance from a trait.

Structure is the active ingredient. The same interview run loosely drops from about .42 to about .19, and a work sample scored by gut feel throws away the consistency that makes scores comparable. The gain comes from every candidate getting the same task, the same questions, and the same rubric.

Work samples are still strong at about .33, and the case for them in technical hiring is real, which is the whole idea behind good coding assessment software. They have one limit worth knowing. Sackett and colleagues note that work samples and job knowledge tests don't suit entry-level hiring when the skills get learned after hire. Our guide to the work sample test covers when to use one and how to score it.

The strongest first round combines the two. Give the candidate a real task, then have a structured conversation about the choices they made. You get the doing and the reasoning on one scale, with the top-ranked method wrapped around a strong one. For that to hold up later the scoring has to be legible, which is why how the scoring and reasoning work matters as much as the task.

Fairness, adverse impact, and the record you can defend

Fairness is not a compliance chore you bolt on after picking a test. The test you pick is the compliance posture. Regulators do not ask whether your assessment felt fair. They ask whether it is job-related and whether it screens out protected groups at a disproportionate rate.

In the United States, the EEOC's guidance on employment tests and selection procedures sets the baseline through the Uniform Guidelines. The common yardstick is the four-fifths rule: if one group's selection rate falls below 80% of the highest group's rate, that is a signal of adverse impact you have to justify with evidence that the test predicts the job.

This is exactly where a generic cognitive battery gets a company into trouble. Job-relevant measures are easier to defend on job-relatedness, though work samples can still show sizable group gaps in Sackett's data, while structured interviews show much smaller ones.

The regulatory layer has grown past the EEOC. NYC Local Law 144 (LL144) requires a bias audit of automated employment decision tools plus candidate notice, and it is now a template other jurisdictions are copying. The EU AI Act classifies most hiring tools as high-risk and expects documentation of how the system works and how it is monitored. Illinois regulates AI use in video interviews through its consent law.

The practical takeaway is simple. Favor an assessment that is job-related on its face and that produces a written record of why each score was given. That record turns a bias audit or a candidate challenge from a scramble into a document you already have.

Our guide to reducing bias in hiring goes deeper on the audit posture. None of this is legal advice; run your program past counsel.

How to run a work sample at first round without booking a human every time

The usual objection to work samples is cost. A senior engineer sitting through a live exercise with every applicant does not scale, so teams fall back to cheap-but-weak proxy tests. That trade-off created the personality-test default in the first place.

You do not have to accept it anymore. Keep the work sample and lose the cost by automating the first round while keeping a human on the decision. In practice that is a structured, role-tuned interview with a live task, scored against a rubric, with a person reviewing before anyone is advanced or rejected.

This is where Expert Hire sits. Expert Hire is not a personality-test or cognitive-aptitude-battery vendor. It runs a structured, conversational AI interview tuned to the role, with a live coding exercise that executes the candidate's real code, scored against a transparent rubric, with a report card a human reviews instead of a black-box pass/fail.

It is a structured-interview-plus-work-sample instrument, built on two of the job-specific methods near the top of the ranking above, not another proxy test. It does not offer personality or aptitude scoring.

The point is not the tool. It is that the scaling excuse for weak tests has expired. You can give every candidate a real task at first round and still keep an engineer out of the loop until the final few.

What to look for when buying pre-employment assessment tools

Once you have decided to assess the actual work, the shopping list for pre-employment assessment tools gets short and specific. Judge a tool against these before you judge it on price.

Reasoning, not a verdict. The output should show why each score was given, with the evidence behind it, not a bare pass, fail, or percentile. A number you cannot explain is a number you cannot defend.
A human in the decision. The tool should support review before any advance or reject. A fully automated reject is the exact thing LL144 and the EU AI Act are built to scrutinize.
An audit trail. You want a durable record of the rubric, the score, and the rationale, exportable for a bias audit or a candidate request.
Job-relatedness on its face. The task should look like the work. If you cannot explain to a candidate why the test maps to the role, a regulator will have the same question.
ATS write-back. The result should land in your existing candidate record automatically, so the assessment is part of the pipeline and not a spreadsheet someone re-keys.
Candidate experience. A test that feels like the job respects the candidate's time and protects your employer brand. A test that feels like a hazing ritual costs you offers.

Run any tool, ours included, against that list. The grounding for how we approach scoring and validity is on our methodology page, written for the buyer who reads the evidence before the marketing.

Frequently asked questions

What is a pre-employment assessment? It is a standardized test given to candidates before an offer to predict how well they will do the job. It can measure aptitude, personality, integrity, or job skills, or it can be a structured interview. The useful ones predict on-the-job performance. The rest just add a step.

What are the main types of pre-employment tests? Cognitive and aptitude tests, personality questionnaires, skills and work-sample tests, integrity tests, and structured interviews. They differ sharply in how well they predict performance. Structured interviews lead the current evidence, with job knowledge tests and work samples close behind; generic personality self-reports sit near the bottom.

Which pre-employment assessment predicts job performance best? Structured interviews, at a mean validity of about .42 in Sackett and colleagues' 2022 re-analysis. Job knowledge tests (about .40), empirically keyed biodata (about .38), and work samples (about .33) follow. They work because they measure job-relevant behavior directly instead of inferring it from a trait or a generic reasoning score. Older sources quoting Schmidt and Hunter's 1998 figures put work samples first, but those estimates were inflated.

Are pre-employment tests legal? Yes, when they are job-related and do not cause unjustified adverse impact. The EEOC's Uniform Guidelines set the US baseline through the four-fifths rule, and newer rules such as NYC Local Law 144 and the EU AI Act add audit and documentation duties for automated tools. Keep a defensible record and consult counsel for your jurisdiction.

Do personality tests work for hiring? As a supporting input, sometimes. As the main assessment, weakly. Traits like conscientiousness carry some predictive value, but self-report is coachable and the validity is low next to a structured interview or a work sample. Use personality data as a tiebreaker, not a foundation.

How can we run work samples without a huge time cost? Automate the first round and keep a human on the decision. A structured, role-tuned interview with a live task, scored on a rubric and reviewed by a person, gives every candidate a real work sample without booking a senior reviewer for each one.

The assessment worth running is the one that measures the work

If you take one thing from this guide, take the ranking. Most pre-employment assessments measure a proxy, and the methods that predict best measure the job directly: a structured interview first, then job knowledge tests and work samples. Personality and aptitude tests can support that decision. They should not be the decision.

So assess the work, keep it structured, keep a human on the call, and keep a record you could hand to an auditor without flinching. If you want to see what that looks like in practice, open a sample report card on the AI interview platform and read the rubric, the transcript, and the reasoning behind each score before you decide whether it holds your bar.

By TK, Growth at Expert Hire. Last updated October 13, 2026. Reviewed by Anand Suresh, CPO at Expert Hire.