What are the alternatives to Adaface for pre-employment skills testing?

Forty minutes, two ways, and the seconds-per-item number nobody prints

Both vendors have settled on forty minutes, and for the same stated reason: completion rates fall as tests get longer. Adaface's science page puts a figure on it — tests of 40 minutes or less have a 75 per cent completion rate against 66 per cent for tests of 41 to 60 minutes. The page gives no citation for those numbers, so a reader cannot check which study or which candidate population they describe, and this page does not repeat them as fact.

What is checkable is what each vendor puts inside the forty minutes. Adaface's cognitive ability test is 20 questions: five verbal reasoning, five numerical reasoning, five abstract reasoning, five critical thinking, at a stated difficulty of Moderate. That is 120 seconds an item. AssessAll's Complete Graduate Aptitude Battery is 40 questions in the same 40 minutes, which is 60 seconds an item, and its speed-calibrated numerical form is 25 items in 22 minutes, or 52.8 seconds.

Two minutes an item on a moderate form is a power measure. The candidate is not under time pressure; what varies between candidates is whether they can do the reasoning, not how fast. One minute an item on a four-construct battery is a paced measure, and part of what it sorts on is working speed. Neither is the better test. They answer different questions, and a buyer who has only been shown a duration cannot tell which one they are buying.

The consequence for reading a score is direct. On a power form, a low score is evidence about reasoning. On a paced form, the same low score could be a candidate who reasoned well and ran out of time, which is why attempted-versus-correct matters on one and barely matters on the other. Ask any vendor, this one included, for the item count and the time limit together, then divide. It takes ten seconds and it tells you what the number on the report is made of.

The validity coefficients on that science page are the pre-2022 ones

Adaface's science page is the most psychometrically literate vendor science page this site has read across eight competitor audits, and the observation that follows is offered in that spirit rather than as a criticism. It is also the most common dated claim in this market, and AssessAll has made the same class of error on its own pages within the last month.

The page publishes five coefficients: general mental ability at .51, work sample tests at .54, unstructured interviews at .38, job experience at .18 and years of education at .10. Those are the Schmidt and Hunter (1998) values, and the page's own headings acknowledge the source — "Rooted in 85 years of research" is that paper's subtitle. The next heading reads "Bringing the research into 2024".

The research moved in 2022. Sackett, Zhang, Berry and Lievens showed that meta-analyses in this tradition had corrected for range restriction too aggressively, and the abstract states that most of the same selection procedures that ranked high in prior summaries remain high in rank, but with mean validity estimates reduced by .10 to .20 points, and that structured interviews emerged as the top-ranked selection procedure. The revised figures reported in the Society for Industrial and Organizational Psychology's own magazine put structured interviews at .42, job knowledge tests at .40, work samples at .33 and cognitive ability at .31.

So a page dated 2024 that argues from .51 and .54 is arguing from numbers its own field revised downward two years earlier, and the direction of the revision matters commercially: a business case built on the older figures is overstated by a knowable amount. The same abstract's closing sentence is that selection procedures remain useful. Nothing here says cognitive testing does not work.

AssessAll's claim to have noticed this is small and should be stated at its real size. This site indexes both papers with the revision attached, and on 28 September 2026 it found a superseded interrater-reliability coefficient still live on the sample-report page a buyer reads, a day after correcting it on three other pages. The honest version of this row is not that one vendor checks its sources and the other does not. It is that coefficient vintage is the thing this market gets wrong by default, and the only defence is a dated correction log that a reader can see.

Adaface publishes its sample items, and that earns the harder question

The cognitive ability page carries complete items with their options — a clock that gains two minutes a day against one that loses two, a critical-thinking item about a pharmaceutical trial's design, an inference item about phone imports and manufacturing jobs. Publishing them is a real act of transparency: "no trick questions" is an unfalsifiable claim until the questions are on the page, and this category is full of the unfalsifiable version.

It also raises a question that applies to every vendor publishing samples, including this one. Are the published items specimens written for the page and never used to score anybody, or are they items from the live bank? The two have opposite consequences. A specimen item costs nothing. A live item published on an indexed page is an item every future candidate can find, which degrades the measure for every client already using it and quietly inflates the scores of the candidates who searched.

AssessAll's position, stated on its own test pages: sample items on this site are written for the page and are never drawn from the live bank or from the seed files, and the pages say so where the items appear. That is a constraint worth publishing rather than hiding, because it converts into the question to ask a vendor. If they will show you live items, what does that imply about the scores they have already reported to you?

Nothing on Adaface's pages says which kind its samples are, and this page does not guess. Ask them, ask us, and ask whoever else is on your shortlist: are these retired specimens or live items, and how often does the bank rotate? A vendor with a good answer will give it in one sentence.

What neither vendor publishes, stated once and applying to both

Across four Adaface pages read on 29 September 2026 — home, pricing, the science page and the cognitive ability test — there is no internal consistency coefficient, no standard error of measurement, no norm-group description, no cut-score guidance and no adverse-impact figure. The science page describes beta testing that keeps only questions which differentiate candidate abilities, which is item discrimination and is the right practice, and it reports no resulting statistic.

AssessAll is in the same position and it is the more awkward half of this page to write. No reliability coefficient and no standard error is published for its aptitude or English forms, so the confidence band around a score on them cannot be computed by a buyer or by this site. No DIF analysis has been run on those banks. The CEFR band boundaries on the English forms are documented in code as a launch calibration rather than as empirically derived cut scores.

That is why the strongest thing either vendor can currently offer a careful buyer is not a coefficient but a specification: what is in the test, how long you get, what a band means, what the score does not license. Both companies publish part of that. Neither publishes the number that would let you put an error bar on a candidate's result, and a page arguing that buyers should demand these numbers would be worthless if it exempted the vendor publishing it.

Who currently answers the query you just typed

Two of the pages ranking for Adaface alternatives were opened on 29 September 2026 before this page was written, on the rule that a comparison you have not checked against the incumbents is a guess. The rank-one result is a competitor's blog listing ten platforms with G2 ratings and feature summaries; it publishes no reliability coefficient, no item count, no time limit and no price for Adaface or for anybody else, and never mentions Adaface's science page. A second vendor page lists fifteen platforms and prints a starting price for fourteen of them — and none for Adaface, the product the page is about, despite Adaface publishing seven tiers openly.

That is the state of the lane, and it is why this page exists in this shape. The information a buyer needs about Adaface is on Adaface's own site, and the pages that rank above it do not carry it. A comparison worth reading is one that opens the vendor's pages, quotes what is on them, dates it, and says which rows go against the author.