Skills assessment vs interviews: what should you screen with?

The short answer

Skills assessments and interviews measure different things, and the strongest hiring processes use both — in the right order. A skills assessment objectively measures whether a candidate can do the work: reasoning, language, judgement, and role-specific ability. An interview explores motivation, context, and mutual fit — things no test can fully capture.

The failure mode most teams live with is using interviews for both jobs. Screening by CV and first-round interviews means capability is judged by impression, which is slow, inconsistent, and biased toward confident presenters. Assess first, interview the qualified — that single change removes most of the noise.

Where interviews fall short as a screen

Unstructured interviews are among the weakest common predictors of job performance, and the research on this is decades deep. Interviewers anchor on first impressions in minutes, weigh fluency and confidence over competence, and rate candidates who resemble themselves more highly — none of which predicts output.

Interviews also do not scale. Thirty first-round conversations cost a hiring team a week of calendar time and still produce scores that cannot be compared, because no two interviews asked the same questions under the same conditions.

What a skills assessment adds

Standardisation is the core advantage: every candidate faces the same items, the same time limits, and the same scoring rules, so a 78 means the same thing for candidate one and candidate three hundred. That comparability is what makes shortlisting defensible.

Modern assessments also widen what can be measured — spoken English scored by AI, situational judgement keyed by experts, typing and numeracy under time pressure, behavioural style — and proctoring keeps remote, unsupervised sittings honest. The result reaches further than any CV: a fresher with no work history can demonstrate exactly the ability the role needs.

How to combine them: assess first, then interview deeper

The pattern that works is a funnel. Use a short assessment battery as the first gate — capability, judgement, and any role-critical skill — and let it cut the pool to the genuinely qualified. Then spend interview time only on that shortlist, and make the interview structured: same questions, defined anchors, scored independently.

Assessment results also make interviews better: a candidate's profile tells the interviewer where to probe — a strong reasoning score but a mid judgement score is an invitation to ask about a real conflict they handled, not a reason to reject.

The correction this page owed you: it is structure, not the interview, that is weak

An earlier version of this guide said interviews are among the weakest common predictors of job performance. That is true of unstructured interviews and it is not true of interviews as a method, and the distinction has become sharper since the field revised its own numbers.

Sackett, Zhang, Berry and Lievens re-analysed the meta-analytic estimates the profession had used since 1998 and found that corrections for range restriction had been applied too aggressively for decades (Journal of Applied Psychology 107, 2022, 2040–2068). On the revised figures the strongest single predictor of job performance is the structured interview at .42, ahead of job knowledge tests at .40, empirically keyed biodata at .38, work samples at .33 and general cognitive ability at .31 — where cognitive ability had been quoted at .51 for a quarter of a century.

So the case for assessing first is not that tests beat interviews. It is that structure beats impression, and an assessment is structure you can buy off the shelf while a structured interview is structure you have to build and maintain — the same questions, defined anchors, independent scoring, every candidate, every time. Teams that fix the screen and leave the interview unstructured have moved the noise rather than removed it.

The practical reading of those numbers is that no single instrument is close to sufficient. The strongest predictor on the list explains under a fifth of the variance in performance on its own, which is an argument for combining methods that measure different things and against treating any one score — an interview verdict included — as the decision.

Running this on AssessAll

AssessAll is built for the assess-first funnel: a catalogue of aptitude, English, typing, judgement, and behavioural assessments, batteries you can compose per role, AI proctoring for remote sittings, and per-candidate pricing with no per-seat licences — so screening 300 applicants is priced like screening, not like enterprise software.

Every candidate's results roll into a Skill Passport with verifiable credential links, so what the funnel measures becomes evidence the hiring manager, and the candidate, can keep.

The fairness question, and which method can actually answer it

Assessments get accused of bias and interviews rarely do, which has the causality backwards. It is not that interviews are fairer; it is that a test produces a number you can audit and an unstructured interview produces a judgement that leaves no auditable trace.

Try it on your own funnel. For an assessment gate you can compute a selection rate per group in minutes and divide by the highest group's rate to get an adverse impact ratio — the adverse impact ratio calculator does it in the browser. For a panel that decided by discussion and recorded a verdict, you can compute the same ratio, but you cannot decompose it: there is no equivalent of a section score, so when the ratio comes in below 0.80 there is nothing to look inside.

That difference is what makes the auditability an argument for testing rather than against it. A structured instrument can be taken apart — per section, per competency, per item — until the gap is localised and fixed. A first impression cannot.

The practical instruction is to run the check on every gate you operate, not only the one with numbers attached. Teams that audit only the assessment are auditing the one stage they can see, then reporting the result as if it covered the process.

There is a newer finding that changes what to do when that audit comes back badly, and it points away from the intuitive fix. Berry, Lievens, Zhang and Sackett re-estimated the personnel-selection meta-analytic matrix (Journal of Applied Psychology 109, 2024, 1611–1634) and report two results together: that the criterion-related validity of general mental ability tests has been considerably overestimated because of inappropriate range-restriction corrections, and that excluding GMA tests from a battery generally has little to no effect on overall validity while substantially decreasing adverse impact. Their conclusion is stated plainly — contrary to popular belief, GMA tests are not a driving factor in the validity–diversity trade-off.

Read that as a pricing correction rather than as permission to stop testing. The standard advice, when a ratio comes in under 0.80, has been to de-weight the ability test and accept a validity cost as the price of a fairer funnel. That advice was priced on validity figures the 2024 re-estimate says were too high, which makes the cost of the change smaller than the received wisdom implies — and it also means the cognitive gate is not automatically the place the problem is. Compute the ratio at every gate, including the interview, before deciding which one to alter.