NeuroFrame and SHL

Side by side

SHL Group Limited (referred to on this page as SHL). SHL and the product names cited here are trademarks of their respective owners; they are used nominatively, to identify the products discussed.

How to read this page

The direct answer first: we are not a neutral observer. NeuroFrame sells an assessment product, and SHL sells assessment products. A page written by one vendor about another is worth reading only if it is built so that you can check it without trusting us. That is the rule this page follows.

The rule has one clause. A statement about SHL appears here only if it can be traced to something SHL published — a product fact sheet, a technical manual, a page on shl.com, a corporate registration record — with a working URL and the date we opened it. You will not find the adjectives "strong", "weak", "shallow" or "sufficient" applied to the other product anywhere on this page, because those are opinions and opinions are not checkable.

Where we write "we could not find", it means exactly that and nothing more: the result of our search of publicly available material on 4 August 2026. It is not a claim that a document does not exist, and it is certainly not a claim about how the company behaves. Product lines change, so every specification here carries the copyright year of the fact sheet it came from (© 2018, © 2020, © 2022, © 2024) alongside the date we verified it.

Numbers about NeuroFrame come from our own single register of published figures — the same file that feeds every page on this site. That register also holds the numbers that do not flatter us. Those are in the section "When to choose SHL rather than us", and they are not a formality.

What SHL is, and who owns it

SHL is a British assessment company. Its own "About" page dates the founding to 1977 and lists the milestones that followed: OPQ in 1984, Verify in 2006, TalentCentral in 2011, acquisition by CEB in 2012, and what the page calls a "relaunch as standalone company" in 2018. The parent entity is SHL Group Limited, registration number 01328744, registered at The Pavilion, 1 Atwell Place, Thames Ditton, Surrey KT7 0NE.

On ownership, the checkable record is the portfolio page of Exponent Private Equity, which we opened on 4 August 2026. It lists SHL under Current Investment, with the transaction dated March 2018, a figure of $400 million, and Fund III. The same page describes the client base as "80% of the FTSE 100 and over 50% of the Fortune Global 500". We found no public announcement of a change of ownership after that, and we make no claim about what may happen next.

The scale figures SHL puts on its own home page are these: "45 million SHL assessments are taken every year, around the world", "over 45 billion data points on workforce skills, performance, and potential", "Providing People Answers to Over 10,000 Businesses Worldwide", and "SHL Integrates with 80+ ATS". These are the company's statements about itself; we did not find a description of how they are counted on the page itself, and we have not verified them independently. We quote them as attributed statements, not as established facts, and we do not dispute them either.

Two things changed recently and are worth knowing if you are evaluating SHL in 2026. First, a distributor — GFB (Getfeedback) — wrote on 13 November 2024 that SHL "will formally cease support" for a set of legacy Verify tests (Verbal, Numerical, Inductive, plus Checking and Mechanical Comprehension) from 1 April 2025, with migration to Verify Interactive and other products. That is a partner's post about a future action, not a statement by SHL, and we do not assert that the migration has in fact completed.

Second, readiness to work with AI. SHL carries an AI Readiness page on shl.com describing what it measures in those terms; as checked on 4 August 2026 that page gives no launch date and no psychometric coefficients. A four-factor model — AI literacy, analytical thinking, continuous learning, AI promotion — referencing "more than one million assessments" is described in a 22 July 2026 post by Josh Bersin. That is a third party's account of the product: we did not find those four factors, that figure or a corresponding dated announcement on SHL's own pages as of the date of checking, and we reproduce them attributed to the post rather than to SHL. A separate checkable fact: SHL and The Josh Bersin Company ran a joint Superworker Skills research programme from September to December 2025, with the terms published on shl.com. We draw no inference from this about the July post.

How the assessment is built: format, task counts, duration

The direct answer first, because this is the single most common misconception in tender comparisons: in the SHL materials we checked as of 4 August 2026 — the fact sheets and product pages listed in our sources — we found no battery of mini-games described and no stated number of games. Three different classes of instrument live under one brand, and it is worth keeping them apart — an adaptive cognitive test with an interactive answer format, work simulations, and forced-choice questionnaires.

Verify Interactive is the first class. Per the G+ fact sheet (© 2018 SHL): allowed time 36 minutes, maximum number of questions 24, one sitting, "Designed for Unsupervised Environment: Yes", question format "Interactive / Activity-based, Adaptive". The separate scales run 18 minutes each: Inductive Reasoning with a maximum of 15 questions and Numerical Reasoning with a maximum of 10, where the candidate's task is to "create charts, graphs, and fill in spreadsheets" (fact sheets © 2022 SHL). SHL's own wording for the format, on its cognitive assessments page, is: "Our immersive assessments upgrade button clicking with 'drag-and-drop' and gamified interactions without compromising professional assessment standards". That page carries no timings — those are in the fact sheets.

Job simulations are the second class, and they are built differently: the candidate works inside a realistic interface. Contact Center Call Simulation (© 2020) — 2 tasks, average testing time 15 minutes, allowed time 20, language English (US). The phone agent simulations (© 2022) — 20 minutes each with 2 scenarios per simulation, and packaged Solutions at 30 and 35 minutes. Conversational Multichat Simulation (© 2024) — average 11 minutes, maximum allowed 25, up to 2 chat tasks. Automata Pro (© 2020) — 2 tasks, average 46 minutes, allowed 60, an IDE environment, "over 40 different programming languages".

The third class is questionnaires, and the distinction matters when you are comparing line items in a proposal. Professional 8.0 — a Job-Focused Assessment — is 16 minutes and up to 76 forced-choice questions (© 2024). The Global Skills Assessment fact sheet (v1.0, 21 May 2024) is 15 minutes and 76 forced-choice triplets covering 96 skills of the Universal Competency Framework, and it says directly that the GSA "is used in all 8.0 Job Focused Assessments (JFAs)" and that it "is not available by itself for stand-alone off the shelf use". OPQ32r is about 25 minutes and 104 questions, untimed. Multitasking Ability (© 2018) is 20 minutes and 38 questions; the fact sheet describes it as a split-screen simulation and gives its Question Format as "Multiple Choice".

We did not come across an off-the-shelf package with a single stated total duration in the published materials: by its own description, SHL assembles a package for the role. The confirmed off-the-shelf combinations we found are the 30-minute Customer Service Phone Solution and the 35-minute Sales & Service Phone Solution. Most of the key products — Verify Interactive G+, GSA, Professional 8.0, Conversational Multichat Simulation, Multitasking Ability, OPQ32r — are marked as designed for an unproctored or unsupervised environment; SHL names adaptive item delivery, a psychometric verification procedure and optional proctoring as the compensating mechanisms.

What the vendor publishes on validity and reliability

The direct answer: for the classic Verify range, SHL published a detailed technical manual, and it contains rather more than validity coefficients. For the Verify Interactive range, we were unable to find a publicly available document of comparable detail as of 4 August 2026. That second sentence is a statement about our search, not about the company.

The SHL Verify Range of Ability Tests Technical Manual, version 2.0, October 2007, reports a meta-analysis: operational validity 0.50 for the verbal test (K = 5 studies, total N = 548, range of observed validities 0.21–0.43) and 0.39 for the numerical test (K = 7, total N = 760, range 0.11–0.34). It also reports a separate criterion study in US retail for a Junior Customer Contact role: N = 89, validity 0.31 for both tests, with internal consistency of the criterion ratings at 0.81.

The same manual publishes reliability and group differences. Internal consistency across 400 generated test forms: verbal 0.81 for managerial/graduate and 0.78 for supervisor/operational levels, averaging 0.80; numerical 0.83 and 0.84, averaging 0.84. Verification tests: 0.77 and 0.79, correlating 0.72 and 0.70 with the corresponding ability tests. Effect sizes (d) by gender, ethnicity and age: verbal 0.06 / 0.11 / 0.04; numerical 0.23 / 0.09 / 0.22. Construct correlations for the Verify Inductive Reasoning test with Raven's APM (0.558) and GMA Abstract (0.538); the study ran on 109 UK university students, of whom 60 also took Raven's APM and 49 took GMA Abstract Form B (manual, p. 37). Item development ran over 36 months with 16,132 participants; the comparison-group sample was 8,436 people.

For Verify Interactive specifically, what we could verify is narrower. The published sample report (© 2019 SHL) names its comparison groups "Interactive Inductive/Numerical/Deductive Reasoning General Composite (INT) v1", and a service page in the same document also lists "Verify G Plus General Population (INT) 2016 comparison group"; the size and composition of those samples are not stated in the report itself. A test review of SHL Verify Interactive G+ (UK English) was published by the British Psychological Society in 2024, DOI 10.53841/bpstest.2024.shlvi. The full text is paywalled; we have not read it, we do not quote it, and we say nothing about what it concludes.

For the simulations and the AI scoring, the fact sheets give timings and lists of scales, and SHL's public AI whitepaper describes methodology. Numeric validity coefficients for the simulations, or agreement rates between AI scoring and human raters, were not something we found in publicly available SHL material on the date of checking.

Regulatory position: audits, obligations, proceedings

Start with the point that most vendor comparisons get wrong. Under New York City's Local Law 144, the duty to commission an annual independent bias audit and publish a summary of it falls on the employer or employment agency, not on the assessment vendor — 6 RCNY §5-301 and §5-303 are addressed to "an employer or employment agency". A vendor that publishes an audit is making a voluntary maturity signal. The presence or absence of such a page on any vendor's site is therefore not a measure of legal compliance, and we draw no conclusion from it about SHL or about anyone else, ourselves included.

What SHL publishes on this itself is its open AI whitepaper (version 1.3, February 2025). It sets out principles for AI use, among them that "No AI assessment should make decisions without human oversight", and it states that SHL audits its AI systems for fairness across gender, ethnicity and age before release, with results "made publicly available for customers". The wording is addressed to customers: availability is described with respect to them. We did not locate those audit reports in open access on 4 August 2026 — a statement about our search, which says nothing about what customers may receive.

On group differences, the classic Verify manual described in the previous section does publish effect sizes by gender, ethnicity and age. For the Verify Interactive range we found no comparable public figures on the date of checking.

The calendar you are actually planning against, as of 4 August 2026: the EU AI Act names recruitment and candidate evaluation as high-risk in Annex III, point 4(a); its Article 50 transparency obligations apply from 2 August 2026, while the Chapter III high-risk regime for Annex III systems was moved to 2 December 2027 by Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026. The Chapter II prohibition on inferring emotions in the workplace has applied since 2 February 2025 and was not softened. In the UAE there is no dedicated law on AI in hiring: the binding norm is Article 18 of the federal PDPL, giving the data subject a right to object to decisions produced by automated processing, and inside the DIFC there is Regulation 10 on autonomous systems.

On complaints and proceedings we make no statement about SHL. Whether a filing exists, and whether it has merit, is established by a regulator or a court — not by a competitor's website.

The reference point: this is where the real difference runs

Both instruments produce a score, but they answer different questions, and the difference is in the reference point rather than in the interface. A percentile norm answers: how does this person look against a comparable population? A role benchmark answers: does this person fall inside the range this specific role requires, in a company at this specific stage of its life?

SHL's reference point is a normative comparison group. The 2007 manual describes 30 verbal and 30 numerical comparison groups built on a 3 (education level) × 4 (industry cluster) scheme, plus a composite norm for each job level, on a sample of 8,436. The Verify Interactive sample report names its groups "Interactive … General Composite (INT) v1" and recommends confirming the result with other methods, including work simulations. We would repeat that same recommendation about our own output.

NeuroFrame's reference point is a role library: 287 professions × 8 corporate lifecycle stages = 2,296 records, built externally — from the Russian Ministry of Labour occupational register and the Adizes lifecycle model — rather than from a sample of a client's own high performers. Two consequences follow. It works from day one, without requiring the client to have dozens of incumbents in a role. And it does not structurally encode the client's existing staff composition into the benchmark.

Why the lifecycle stage carries weight is a computation on those 2,296 records: for 99% of the 287 professions, the requirement ranges at "Infancy" and at "Bureaucracy" overlap by less than half, with a median range overlap of 0.40. Of the 153 non-overlaps between the extreme stages, 152 fall on risk appetite. The travel of the range midpoint across the lifecycle is 18.8 percentile points for risk appetite, 15.4 for openness to new, 11.5 for conscientiousness and 9.0 for cognition; result focus barely moves at all.

Population comparison and benchmark comparison do not exclude each other, and NeuroFrame also has a comparison sample: 14,850 people in real jobs, drawn across 500 companies, 21 industries, 21 functional areas and 8 grades. It is smaller than the normative bases of instruments with forty years of history, and we say so plainly rather than around it.

How the NeuroFrame method differs — on substance

NeuroFrame is one continuous simulation lasting 30 to 60 minutes. Not a battery, not a set of named mini-games, not a questionnaire. The person runs a situation in which decisions carry delayed consequences, and eight parameters are read from behaviour — three in the cognition domain (Mental Efficiency, Learning Agility, Progress Monitoring) and five in the personality domain (Openness to New, Risk Appetite, Result Focus, Agreeableness, Conscientiousness).

Why one long session rather than a set of short ones. Dynamic simulations with delayed consequences measure complex problem solving — a construct that overlaps with general cognitive ability by roughly 18% of variance (r ≈ .43 in a meta-analysis of 47 studies). Part of what is being observed is therefore not what a classical ability test picks up. Separately, the decline of sustained attention as time on task grows — vigilance decrement — is one of the most replicable effects in cognitive psychology at the group level, and a long session lets you observe behaviour both fresh and fatigued. We do not turn that into an individual "accumulated fatigue" score: difference scores have poor test–retest reliability, and we will not publish a number we cannot defend.

What we do not claim is as load-bearing as what we do. A game format is not more accurate than questionnaires: a systematic review of 34 studies (Ramos-Villagrasa et al., 2022) concludes that the game-based format does not offer advantages sufficient to recommend it in place of conventional methods — with the exception of markedly better candidate reactions. A long simulation is not, by itself, higher in predictive validity either: on the Sackett, Zhang, Berry and Lievens (2022) re-estimation, structured interviews sit at .42, job knowledge tests at .40, work samples at .33, cognitive ability tests at .31 and assessment centres at .29. Assessment does not replace the interview; it prepares it. And the game format reduces but does not eliminate deliberate faking (controlled experiment, N = 171).

Where the method genuinely differs is in the justification it carries. Each of the 2,296 library cells holds a human-readable explanation — 2,296 unique texts, 1.07 million characters, 464 characters per cell on average, of which 32% name the rule that fired outright. In 303 cells of the 2,296 (13%) the benchmark permits a high risk appetite alongside a low floor on cognition or on progress monitoring — meaning the method can say "fits on the parameters and still warrants attention". A single aggregate match percentage cannot say that in principle.

Anonymity is architectural rather than procedural. The client receives codes and maps them to people itself; NeuroFrame receives no name, no gender and no age — those attributes are simply absent from the input, so the model does not train on them directly. That is not proof that no indirect association exists: we have no formal adverse impact ratio report yet, and we say so in "When to choose SHL rather than us". The flip side of the same design is that we have no ATS connectors — and that is a direct consequence, not an oversight we intend to argue away.

Languages and presence in MENA

If you are buying in the UAE or Saudi Arabia, the practical question comes before the psychometric one: in what language does the candidate actually sit the assessment, and is there a legal entity in the region to contract with?

SHL's regional presence is documented in its own records. Registered entities include SHL Middle East and Africa FZ-LLC at Dubai Knowledge Village, Block 2A Suite G46, registration number 18866, and SHL Product Limited Abu Dhabi Branch in Musaffah. The global offices page lists Riyadh among its Middle East & Africa locations and describes the network as "offices and distributors in over 40 locations across the world". The site also carries a separate English (MENA) regional variant.

On languages, here is what we could verify and no more. The catalogue card for SHL Verify Interactive – Deductive Reasoning lists Arabic among its localisations, alongside more than twenty others including Chinese, Czech, Danish, Dutch, English (USA and International), Finnish, French, German, Greek, Hungarian, Indonesian, Italian, Japanese, Korean, Latin American Spanish, Norwegian, Polish and Portuguese. The contact centre simulation fact sheets (© 2020, © 2022) state English (US), and the Automata Pro fact sheet (© 2020) states English. We do not have SHL's full language register and we make no claim that other localisations do not exist.

On our side the limitation is straightforward and worth stating before you ask: the NeuroFrame client role library exists in Russian only as of this page's date. For an English- or Arabic-language candidate flow that is a constraint today, and it belongs in the next section rather than in a footnote.

Which question are you actually solving

Most of the confusion in these comparisons comes from putting three different jobs into one column of a spreadsheet. Separating them makes the choice almost mechanical.

If the job is to screen a large flow on reasoning ability in 18 to 36 minutes, with adaptive item delivery and percentile norms, that is an ability-test job. Verify Interactive is built for it, and our continuous session is plainly longer than that job needs.

If the job is to check a specific work skill inside a realistic interface — a call, a chat queue, code in an IDE — that is a work-sample job, and SHL's job simulations are built for it, with the language constraints noted above.

If the job is to see how a person behaves under uncertainty when consequences are delayed, and to set that against what a specific role requires at your company's current lifecycle stage, that is what NeuroFrame is built for. Not better, differently: a different construct, a different reference point, a different output.

One more distinction that belongs on the procurement side rather than the psychometric one. A score that screens candidates out is regulated more tightly, in every jurisdiction discussed above, than a score that opens a conversation. Decide first which of the two you are buying — then compare instruments. And treat all ranges accordingly: a gap on one or two parameters is something to explore at interview, not grounds for rejection.

Questions

Is SHL a game-based assessment?

Not in the sense of a battery of neuro mini-games. In the SHL materials we checked, no "number of games" is stated. The word "gamified" refers to the answer format inside a cognitive test: its cognitive assessments page describes "immersive assessments upgrade button clicking with 'drag-and-drop' and gamified interactions". Underneath is a classical adaptive reasoning test — G+ is 36 minutes and a maximum of 24 questions in one sitting (fact sheet © 2018 SHL). The job simulations — contact centre, multichat, coding in an IDE — are a separate and differently built class. (Sources for these figures are listed under Sources below: SHL's Cognitive Assessments page, the Verify Interactive G+ fact sheet © 2018 SHL, and the Skills Assessments and Job Simulations page.)

How long does the assessment take — with SHL, and with you?

With SHL it depends on the assembled package: Verify Interactive G+ is 36 minutes, the separate Inductive and Numerical scales 18 each, Professional 8.0 is 16, and the simulations run from an average of 11 minutes (multichat) to an average of 46 (Automata Pro). We found no published total package duration in SHL's public materials as of 4 August 2026; the confirmed off-the-shelf bundles are 30 and 35 minutes. NeuroFrame is 30–60 minutes in one continuous session, always structured the same way. If you need a fifteen-minute screen for a high-volume flow, our session is longer than you need. (Sources for these figures are listed under Sources below: the SHL fact sheets © 2018, © 2020, © 2022 and © 2024.)

Which of you publishes more validity data?

The honest answer splits in two. For the classic Verify range SHL has a detailed 2007 technical manual: a meta-analysis (0.50 verbal, K = 5, N = 548; 0.39 numerical, K = 7, N = 760), reliability coefficients, construct correlations and effect sizes by gender, ethnicity and age. For the Verify Interactive range and the simulations we found no publicly available document of comparable detail as of 4 August 2026 — a statement about our search. NeuroFrame has no peer-reviewed publication of its own validation studies at all, and we say so plainly. The two vendors' numbers cannot be placed in one column: they come from different designs and different samples. (The source for the SHL figures is listed under Sources below: SHL Verify Range of Ability Tests, Technical Manual v2.0, October 2007.)

Does SHL have a published bias audit, and is it required to publish one?

Under New York City's Local Law 144 it is the employer or employment agency, not the assessment vendor, that must publish a summary of an independent bias audit — 6 RCNY §5-301 and §5-303 are addressed to the employer. So the presence or absence of such a page on a vendor's site is not a measure of compliance with LL 144, and we draw no conclusion from it — about SHL or about ourselves. SHL itself states, in its open whitepaper (version 1.3, February 2025), that it audits its AI systems for fairness across gender, ethnicity and age before release and that the results are made available for customers. We did not locate those reports in open access on 4 August 2026 — which speaks only to our search. NeuroFrame has no formal bias audit at all; ask every vendor for the document, us included. (Sources are listed under Sources below: the DCWP Notice of Adoption, 6 RCNY §§ 5-300–5-304, and the whitepaper "How SHL is using AI Today", version 1.3.)

Can we use both instruments together?

Yes, and that is usually more sensible than choosing. They cover different constructs: an adaptive ability test answers "how does this person reason against a comparable population", a work sample answers "can they do this specific job", and a continuous simulation answers "how do they behave under uncertainty with delayed consequences, and does that fall inside the role's range at our current lifecycle stage". The Verify Interactive sample report itself recommends confirming the result with other methods, including work simulations. We say the same about our own output: on the Sackett et al. (2022) re-estimation the structured interview (.42) still sits above every method named here, and assessment does not replace it. (Sources are listed under Sources below: the Verify Interactive sample report © 2019 SHL and Sackett, Zhang, Berry & Lievens, 2022.)

We need Arabic. What is actually available?

On SHL's side, the catalogue card for Verify Interactive – Deductive Reasoning lists Arabic among its localisations (checked 4 August 2026). At the same time the contact centre simulation fact sheets (© 2020, © 2022) state English (US) and Automata Pro (© 2020) states English. We do not have SHL's full language register, so we assert nothing about localisations we did not see — ask the vendor directly about the specific product. On our side, the NeuroFrame client role library currently exists in Russian only. If your candidate flow is Arabic-speaking, that is our limitation, not our advantage. (Sources are listed under Sources below: the Verify Interactive – Deductive Reasoning catalogue card and the SHL fact sheets © 2020 and © 2022.)

We need verbal and numerical ability as separate scales. Do you cover that?

No. NeuroFrame's single continuous simulation yields eight parameters — three cognitive, five personality — and verbal and numerical reasoning are not among them. That is an ability-test job, and the Verify Interactive line is built precisely for it: Inductive is 18 minutes and a maximum of 15 questions, Numerical 18 minutes and a maximum of 10, G+ 36 minutes and a maximum of 24. If your tender requires those specific scales, we are not a fit — and it is better to establish that on the first call than on the third. (Sources for these figures are listed under Sources below: the Verify Interactive G+ fact sheet © 2018 SHL and the Inductive and Numerical fact sheets © 2022 SHL.)