Skip to main content
Systematic Hiring

How to pick the right recruitment assessment tool for your hiring needs

What to require from a recruitment assessment tool: predictive validity, independent certification, candidate data in the EU, and where AI actually belongs.

Reading time: 12 minutes

You've been asked to pick a recruitment assessment tool. Every shortlist looks the same on paper: validated, fair, quick for candidates, popular with hiring managers. Almost none of them publish the one number that would let you tell them apart.

This guide covers what to ask for, what the research actually supports, and where most evaluations go wrong.

What a recruitment assessment tool is actually for

A good assessment predicts job performance better than a CV, an unstructured interview, or years of experience. That's the whole job. Everything else is packaging.

The reference point is Sackett, Zhang, Berry and Lievens (2022), the largest re-analysis of selection research to date. It recalculated the numbers the field had relied on for 20 years, including the ones Hunter and Schmidt produced, and found that a standard statistical correction had been applied in a way that inflated them.

The ranking mostly held. The numbers didn't: average validity estimates came down by between .10 and .30. Structured interviews came out top of the list.

So the two findings that should shape your process: structure beats instinct, and years of experience and years of education still predict job performance poorly.

Assessments themselves aren't new. Imperial China used them to select civil servants, and both the US and Europe used cognitive and personality testing for military selection during the First World War. What's changed is how much evidence now sits behind the method, and how little of it most buying processes examine.

Which assessment answers which question

Four types cover most hiring, and they answer different questions.

Logic test

Measures general cognitive ability: how quickly someone learns and solves problems they haven't seen before. One of the strongest predictors across role types, and the most useful input you have when you're hiring for potential rather than for a finished skill set.

Alva's logic test is a non-verbal matrix reasoning test, a format used widely in both research and practice. The format minimises the role of previous knowledge, so what it captures is pattern recognition rather than schooling.

Where it falls short: practice improves scores, and some candidates, including neurodivergent candidates, experience timed tests differently. Treat it as one input, never the verdict.

Personality test

Measures stable traits that shape how someone works with others and handles pressure. The Big Five model is the gold standard here, with decades of research behind it. Useful for role fit and for preparing the interview. Weak on its own, strong in combination.

Where it falls short: people answer with what they think you want to hear. Some candidates find personality questions intrusive, so the quality of the candidate-facing experience matters more here than anywhere else.

Coding test

Measures demonstrated skill for engineering roles. Narrow by design, so treat it as evidence about one dimension. A candidate might code in a different language to the one you use and still be the fastest learner in the pipeline.

Structured interviews

The same questions, the same criteria, the same scoring, for every candidate. Structure is what separates an interview that predicts from an interview that confirms a first impression, and it's what put the structured interview at the top of Sackett and colleagues' ranking.

The combination matters more than any single part. A logic test, a personality test and a structured interview, all scored against criteria set before anyone applied, is a stronger process than any one of them run alone.

Where does AI actually belong in candidate assessment?

This is the question most buyers ask too vaguely, and the one where the vendor answers diverge most.

There are two designs being sold under the same label.

In the first, a model scores the candidate. It reads a video interview, a written answer or a game, and produces a rating. The model is the measurement.

In the second, a validated psychometric instrument produces the score, and a model works on top of it: reading the score against the criteria set for the role, explaining what it means, drafting the interview questions, and leaving the decision to a person who can see the evidence.

Both get marketed as AI-powered pre-employment assessment platforms. Ask which one is in front of you, and ask early, because the answer decides what you'll have to document later.

Three questions get you most of the way there:

  • What exactly produces the score, and has that instrument been validated independently?
  • Can a human see the evidence behind every recommendation, or only the recommendation?
  • Is anything decided without a person in the loop?

A vendor who can't draw you that line on a whiteboard in two minutes hasn't thought about it carefully enough to sell it to your legal team.

What the EU AI Act requires, and when

AI used in recruitment and candidate selection sits in the AI Act's high-risk category. The obligations that come with it, including risk management, data quality, technical documentation, candidate transparency and meaningful human oversight, were originally due to apply from 2 August 2026.

That date moved. The Digital Omnibus, which entered into force on 27 July 2026, pushed the high-risk obligations for employment and recruitment systems to 2 December 2027, with no conditions attached.

Thirteen months sounds comfortable. It isn't, if the documentation you need has to come from a vendor who hasn't started writing it.

Some of the AI Act already applies today. The prohibited practices in Article 5 have been in force since February 2025, and that list includes emotion recognition in the workplace. The AI literacy duty has applied since the same date. The transparency duties in Article 50, which cover telling people when they're interacting with an AI system, applied from August 2026. GDPR Article 22, on decisions made by automated processing alone, has applied since 2018 and never went anywhere.

So the practical read for a buying process running now: you're procuring against a 2027 deadline, using evidence the vendor has to produce, while a handful of rules already bind you today. Ask for the documentation during the evaluation, not after the contract.

A vendor certified to ISO/IEC 42001, the management standard for AI systems, has done part of this work already. Certification isn't compliance, and no vendor makes you compliant on its own. It does tell you someone other than the vendor has audited them.

New call-to-action

Seven things to require from a vendor

Ask for all of them in writing.

1. Published predictive validity for the roles you actually hire. Not a general claim about assessments. A number for role families close to yours, and the study behind it. If a vendor can't show it, the burden sits with them.

2. Independent certification, not self-certification. A vendor's own white paper is marketing. A third-party audit is evidence. Ask who certified the tests, when, and against which standard. ISO 10667 covers the assessment process, ISO/IEC 42001 covers AI management, ISO 27001 covers information security. They're three different things, and holding one is not holding the others.

3. Adverse impact testing. Ask how the assessment performs across gender, age and background, and ask to see it. A fairness claim with no measurement behind it is a hope.

4. Candidate completion and satisfaction, measured across all customers. Every candidate you lose at the assessment step is a candidate your competitor gets to meet. Ask for the figure across the whole customer base rather than a reference account.

5. Where candidate data lives, and who touches it. For European hiring, ask where the data is stored, which sub-processors are involved, what the retention period is, and whether candidate data is used to train models. Get the answers in the DPA, not in an email.

6. Exactly where AI acts, and where a person decides. See above. This is the one that will come back to you.

7. What year two looks like. Most assessment programmes fail on adoption, not on science. Ask who sets the system up around your roles, who trains your hiring managers, and what the vendor is measured on after go-live.

How to check the science without a psychometrician on staff

Three checks you can run yourself.

Ask for the technical manual. Reputable vendors have one and will send it. Its absence is itself an answer.

Check each certification against the standard it claims, using the list above. Vendors are not always precise about which one they hold.

Ask which meta-analysis the validity claims rest on, and when they were last revised. If the answer is still Schmidt and Hunter (1998) with the original figures attached, the vendor hasn't updated its homework since those figures came down by .10 to .30.

What this looks like in practice

Swedish watch brand Daniel Wellington had the problem most consumer brands would like to have: a steady flood of applications for every open role. For the talent acquisition team that meant hours of CV screening to surface a handful of names.

Moving the assessment to the front of the process redirected that time to the interview stage, where hiring managers could judge role fit against criteria instead of guessing from a CV.

Convenience and fuel chain Circle K went the same route for a different reason. Talent Acquisition Specialist Martina Puranen: "The Alva platform is very userfriendly which we really appreciate. It's a huge timesave for us, our old test took about 4-5 hours to administer per candidate, here we can test a much larger number of candidates in a much faster time and work to be less bias."

Where Alva fits

Alva is a hiring system built on validated assessment. The logic and personality tests are independently certified by DNV, and the company holds ISO 10667 for the assessment process, ISO/IEC 42001 for AI management and ISO 27001 for information security. Nine years of work sit behind it, with around 650 customers and more than 5 million candidates assessed.

The division of labour is the part worth understanding. The assessment produces the score. The agent reads that score against the criteria set for the role, explains what it means, and drafts the interview material for the hiring manager to review. A person makes the decision, with the evidence in view.

Each test takes about 15 minutes, and difficulty adapts as the candidate answers, so fewer questions are needed for a precise result. Candidates get the same full report the recruiter sees, keep it, and can reuse it for other applications.

Customers report time to hire falling 60%, from 50 days to 20, and 73% report fewer low-performing hires. Candidate completion sits at 91%, with 81% satisfaction across all assessments.

Seven questions to ask before you sign

  1. What's your predictive validity for roles like ours, and who verified it?
  2. Which standard are you certified against, by whom, and when?
  3. How does the assessment perform across gender, age and background?
  4. What's your candidate completion rate across all customers?
  5. Where is candidate data stored, which sub-processors touch it, and is it used to train models?
  6. Where exactly does AI act, and where does a person decide?
  7. Who sets this up around our roles, and what are they measured on after go-live?

A vendor who answers all seven without hedging is worth shortlisting. A vendor who answers two has told you something useful.

Frequently asked questions

What is a pre-employment assessment tool?

A pre-employment assessment tool measures a candidate's ability, personality or skills before a hiring decision is made, so the decision rests on evidence about the person rather than on a CV. The common types are cognitive or logic tests, personality tests, skills and coding tests, and structured interviews.

What is an AI-powered pre-employment assessment platform?

The term covers two different designs. In one, a model scores the candidate directly. In the other, a validated psychometric instrument produces the score and a model explains it in role context while a person decides. Ask a vendor which design you're buying, because the answer determines your documentation and oversight obligations under the EU AI Act.

Are AI hiring assessments allowed under the EU AI Act?

Yes, with conditions. Recruitment AI is classified as high-risk, and the obligations that come with that classification apply from 2 December 2027, moved from 2 August 2026 by the Digital Omnibus. Some rules already apply: the Article 5 prohibitions, which include emotion recognition in the workplace, and the Article 50 transparency duties. Compliance depends on how you deploy the system, not only on which vendor you pick.

Which assessment best predicts job performance?

In Sackett, Zhang, Berry and Lievens (2022), the structured interview came out as the top-ranked selection procedure, with cognitive ability tests also high in the ranking. No single method is the answer on its own. A logic test combined with a personality test and a structured interview, scored against criteria set in advance, predicts better than any one of them alone.

How long should a candidate assessment take?

Alva's logic and personality tests take about 15 minutes each and adapt question difficulty as the candidate answers. Any assessment that runs over an hour should come with an explanation of what the extra time buys you.

Do candidates see their own results?

With Alva, yes. Every candidate gets the same full report the recruiter sees, keeps it, and can reuse it. Many platforms don't share results with candidates at all, so ask.

Can psychometric testing platforms connect to our ATS?

Most do. Alva integrates with Teamtailor, Greenhouse, Lever and others, and can also run as the hiring system itself in place of an ATS.

What is predictive validity, and why does it matter?

It measures how well an assessment score forecasts actual job performance. It's the one number that tells you whether an assessment is adding value or just adding a step. Ask for it per role family, with the study attached.

Find the right assessment tool for your hiring

Hiring the best people stops being a question of gut feeling once the evidence is on the table in front of the people making the call.

See how it works on the roles you're hiring for right now.

New call-to-action

See how Alva works for your team.

Every team is different. Let us show you exactly how Alva works for yours.