NeuroFrame and Aon Assessment (cut-e)

Side by side

Aon Assessment Solutions (a business of Aon plc; the acquired cut-e GmbH was folded into it in 2017)

What Aon Assessment (cut-e) is

Aon Assessment is the line of psychometric tests and gamified modules that Aon plc (NYSE: AON) has developed since acquiring the German firm cut-e GmbH. The acquisition was announced on 4 May 2017; financial terms were not disclosed, and cut-e was headquartered in Hamburg. The press release does not give a closing date, and Aon's own material puts it loosely: “In May 2017”. cut-e was founded in 2002.

At the time of the deal cut-e reported assessing 12 million candidates a year in more than 70 countries and more than 40 languages, from 35 offices. Its founders — Andreas Lohff, Achim Preuss, David Barrett and Espen Skorstad — joined the leadership of the combined group. Aon's own wording for the merged brand is “cut-e and Aon, as Aon Assessment Solutions”. The legal entity named in the ACLU complaint to the US Federal Trade Commission is Aon Consulting, Inc.

As of 4 August 2026 the products sit on aon.com under Talent and Rewards. When we requested assessment.aon.com that day, it returned a 301 redirect to aon.com/en/capabilities/talent-and-rewards. We did not encounter the smartPredict brand on the live candidate page or in the live Talent Assessment Products and Tools catalogue; the Challenge tests themselves are listed on the candidate page, and the Pre-Hire Talent Assessment page carries a client quote about Aon's “gamified assessments” — 70 per cent positive candidate feedback in a Deloitte case. We record only what we saw, and we draw no conclusion about what changed in Aon's product line, or why.

The launch of the People Insights platform was reported on 11 April 2025; the description of that platform, explicitly aimed in part at the Middle East, names psychometrics, day-in-the-life simulations and competencies such as agility and learnability. Gamification is not mentioned in it. We report this as a published item in a trade outlet, not as a statement about the state of Aon's product line.

Scale, in the vendor's own recurring wording on live pages: “more than 30 million assessments in over 40 languages across 90 countries” each year. A 2019 Aon presentation carried a related set of figures — 30 million assessments, 120 countries, 40 languages, 40+ years of experience and 2,000 staff in Global Talent.

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

The short answer: it is a set of separate short modules, each on its own timer, rather than one continuous session. On the open pages we checked we found no total battery duration — the employer chooses which modules to combine.

As of 4 August 2026, Aon's official candidate page lists these durations: gridChallenge — 9 minutes (nine items of increasing difficulty), motionChallenge — 6 minutes, digitChallenge — 6 minutes, gapChallenge — 6 minutes, switchChallenge — 6 minutes, chatAssess — 20 minutes. The constructs Aon names on the same page: working memory, complex planning, deductive logical reasoning, numeracy and situational judgement. The ADEPT-15 personality questionnaire is a separate instrument; per model cards published by the ACLU and updated on 3 October 2024, it runs about 25–30 minutes.

The composition of the line has been described differently over time. In Aon's own presentation at the GBA Workshop (Minneapolis, August 2019) the smartPredict series is introduced as “Choose among four challenges in the suite” — motionChallenge, gridChallenge, switchChallenge, digitChallenge. The current candidate page lists five Challenge tests plus chatAssess.

Adding up the four classic modules at the vendor's published timings (9 + 6 + 6 + 6) gives 27 minutes. That arithmetic is ours, not Aon's: a real session at a given employer may be shorter, or considerably longer — for instance once chatAssess and ADEPT-15 are added. One discrepancy is worth recording: the prep vendor JobTestPrep lists digitChallenge at 5 minutes against 6 minutes on Aon's own page; the other timings agree.

On positioning, the ACLU quotes Aon's documentation directly: the test is “appropriate for use early in the recruitment process, to screen out applicants prior to more resource-intensive hurdles”. The same materials record a gridChallenge score scale of 0–100 and version 3.0, dated 22 December 2022.

What the vendor publishes on validity and reliability

The main open vendor document we were able to obtain is Alina Siemsen's (Aon's Assessment Solutions) presentation “smartPredict — Development Insights and Study Results of Aon's Gamified Assessment Series”, GBA Workshop, Minneapolis, August 2019. We describe its contents because it is open and checkable; that does not imply other documents do not exist: per the ACLU's description, Aon's gridChallenge technical documentation runs to more than 70 pages and contains reliability and validity sections — we were unable to open those documents during our check, so we do not characterise their contents.

Construct equivalence. The presentation reports the relationship between gridChallenge and its predecessor G.A.M.E.: r = .71 (p = .000) overall, r = .74 and r = .69 by presentation condition, N = 306 participants with complete data (68 per cent of those recruited), sampled through Amazon Mechanical Turk, study run in May–June 2019. The stated conclusion is verbatim: “Results prove that G.A.M.E. and gridChallenge measure the same construct”. The same document reports no difference between mobile and desktop administration of gridChallenge: Cohen's d = −.12, N = 350.

Candidate reactions. A desktop/mobile comparison across seven reaction scales for the four challenges was run on N = 374 (Amazon Mechanical Turk), with a separate N = 211 for digitChallenge under revised instructions. The largest number of significant differences (p < .05) in the table falls on motionChallenge — on the Engagement, Effort and Ability to Improve scales.

Group differences. The same document contains Aon's own Adverse Impact section: by sex (men — women) d = −.05, by age (under 40 versus 40 and over) d = −.02, and by ethnicity (White — Asian — Black — Hispanic — two or more) from −.35 to −.48. We reproduce these as the vendor publishes them, without interpretation.

Separately there is a cut-e/Aon marketing piece on chatAssess at Rolls-Royce (document code C1016INT 10.17, i.e. around October 2017): “65 per cent more likely to perform highly in the Assessment Centre”, “50 per cent more likely to get a job”, completion rising from 81 and 74 per cent to 95–98 per cent, 9,309 registrations and 9,138 completions. It is a vendor marketing piece; we found no methodology, study design or statistics reported in it, so we cite it as a published 2017 vendor claim rather than as a source of validity evidence.

Finally, the wider format context. We were unable to find independent peer-reviewed research on Aon's gamified products specifically in our search of open sources. The peer-reviewed work that exists addresses game-based assessment as a class: a meta-analysis of GBA convergent validity against conventional self-report personality measures reports r = .516 (Z = 8.088, p < .001) across 18 studies from 13 articles published 2002–2025 (International Journal of Serious Games, 30 October 2025), and Ohlms et al. (Applied Psychology, October 2025, N = 156 and N = 92) find that video-game experience relates positively to cognitive GBA scores and negatively to academic performance. The test vendor is not named in the open abstract, and the study is not attributed to Aon's products.

What the public documents show: rules, complaints, filings

Start with the point that clears up half the confusion. Under New York City's rules implementing Local Law 144, the duty to publish an annual independent bias audit falls on the employer or employment agency, not on the assessment vendor (6 RCNY § 5-301 and § 5-303). We did not find a separately published bias audit for Aon's products in open sources. That states the result of our search: it is neither a claim that no such document exists nor any conclusion about anyone's compliance with the law.

The FTC complaint. In late May 2024 the American Civil Liberties Union filed a 50-page complaint, “In the Matter of Aon Consulting, Inc.”, with the US Federal Trade Commission under Section 5 of the FTC Act. It concerns ADEPT-15, vidAssess-AI and gridChallenge. In October 2024 the ACLU additionally published model cards for those three tools. We found no publicly announced FTC decision as of 4 August 2026. A complaint is the filer's position; there is no regulator's finding on it, and we draw none.

What the filer asserts. Citing Aon's technical documentation for gridChallenge (p. 57, table 43), the ACLU writes that Black participants in the study sample scored on average .48 standard deviations below White participants; citing switchChallenge documentation dated 4 September 2023 (p. 34, table 17), it reports a figure of 1.21 standard deviations — and the ACLU itself specifies that this refers to one study sample. These figures cannot be generalised to the product as a whole, and we do not generalise them. A separate strand of the complaint concerns criterion-related validity: the ACLU states that across more than 70 pages of gridChallenge documentation the criterion-related validation described is that of the predecessor G.A.M.E., while immediately noting that it is unclear to the organisation whether such testing was conducted for gridChallenge itself, and that it “takes no position” on whether Aon's validation work is otherwise sufficient.

EEOC. Per the ACLU's October 2024 materials, charges were filed with the Equal Employment Opportunity Commission concerning an unnamed mid-sized US-headquartered employer that used gridChallenge together with ADEPT-15, alleging race and disability discrimination. We found no public outcome as of the publication date. The trade outlet HR Dive reported in June 2024 only that a similar complaint was pending with the EEOC, without a filing date. In the same materials the ACLU notes that the ADEPT-15 technical documentation it had relied on was, to the organisation's knowledge, no longer publicly available at the time of writing.

The frame all of this sits in. New York's rules set a minimum transparency format — selection rates and impact ratios by sex, race and their intersections, the number of candidates with unknown demographics, and the audit date; an auditor connected to the tool's development, to employment with the parties, or by financial interest does not count as independent. In the European Union, AI systems used to evaluate candidates in recruitment are classed as high-risk (Annex III, point 4(a) of Regulation (EU) 2024/1689); Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026, moved the application of Chapter III obligations for Annex III systems to 2 December 2027, while the prohibition on inferring emotions in the workplace (Art. 5(1)(f)) has applied since 2 February 2025 and was not postponed. The UAE has no dedicated AI-in-hiring statute: the binding provision is Article 18 of the federal PDPL, the right to object to a decision made by automated processing, and the DIFC operates Regulation 10 on autonomous systems. We describe requirements and link to sources; we offer no legal conclusions and do not assert that any vendor is in breach of them.

Languages and presence in MENA

For a buyer in the Gulf this is often the deciding item, so it deserves separate and careful treatment. Aon's official wording, unchanged since 2017 and present on live pages, is: “more than 30 million assessments in over 40 languages across 90 countries”. On the vendor pages we checked, we did not find a list of the specific languages, nor information on which of them cover the gamified modules in particular. That is the result of our check, not a claim that such versions do not exist: we could neither confirm nor rule out Arabic localisation of the challenges from open sources, and the right way to settle it is to ask the vendor directly.

Aon's regional presence is documented in open sources; on that record it is primarily consulting-led. The UAE office is at the Dubai International Financial Centre, Currency House Tower 2, Level 5; the location page names Risk Capital and Human Capital, including talent and rewards. That page carries no Arabic-language content. An industry profile citing humancapital.aon.com reports regional operations headquartered in the UAE and Saudi Arabia, 15+ years across 30 countries, more than 70 specialist consultants and over 1,000 regional clients.

The launch of People Insights, explicitly positioned for the Middle East, was reported on 11 April 2025; the same piece puts the global talent assessment market at roughly 25 billion dollars.

NeuroFrame's position here is simpler and, frankly, thinner: the instrument exists in Russian and English, the client-facing role profile handbook exists only in Russian, and there is no Arabic version. If your candidate pool speaks and tests in Arabic, that is our weakness, not our strength, and we will not pretend language is a secondary question.

The regional context raises the stakes for both sides. In Saudi Arabia, per the General Authority for Statistics (GASTAT) for Q1 2026, unemployment among nationals fell to 6.4 per cent, and to 9 per cent among Saudi women. That increases demand precisely for high-volume selection of local candidates — which means instruments available to the candidate in the candidate's own language.

The norming model: what a result is compared against

A norming model answers the question “high relative to what”. It determines what the instrument actually reproduces: an externally specified requirement for the role, or the statistics of the people already employed. That choice drives both whether the tool works from day one and what it structurally encodes.

What is known about Aon. We did not find a public description of the norming model for the gamified modules on the vendor pages we checked. The model cards published by the ACLU and updated on 3 October 2024 give gridChallenge version 3.0, dated 22 December 2022, and a 0–100 score scale, and describe the instrument as “fast-paced mini games” positioned for early screening. Open sources took us no further, and we will not fill the gap with guesswork: this is a question for the vendor.

What is known about NeuroFrame — because it is our product and we owe you the whole of it. The requirement benchmark is built externally: from the Russian Ministry of Labour's occupational register and professional standards, and from the Adizes corporate lifecycle model. 287 professions × 8 stages = 2,296 cells, each a range across eight parameters. A candidate's score is compared against a reference sample of 14,850 people in real jobs: 21 industries, 21 functions, 8 grades, 500 companies.

One direct consequence of building the benchmark externally: the method does not require the client to already employ dozens of people in the role — it works from day one. And it does not structurally encode the existing workforce, because it was not trained on it. A second consequence is architectural: the client sends us anonymised codes and maps them to people itself. Under this arrangement we do not receive a name, a sex or an age, so a model cannot be trained directly on protected attributes. That is not the same as an absence of bias: proxies for protected attributes may still be present in behavioural data, and we have no formal adverse impact ratio report yet — we say so in the “When not to choose us” section.

We will not rank the two norming models against each other: we have no Aon-published data on its own norming, and comparing on a parameter for which one side has published nothing is rhetoric, not comparison.

How the NeuroFrame approach differs

The difference between the instruments is one of measurement genre. Aon Assessment is a modular battery of short tasks, each with its own construct and its own timer, assembled by the employer to fit the job. NeuroFrame is a single continuous strategic simulation of 30–60 minutes, from the behaviour inside which eight parameters are read: three cognitive (Mental Efficiency, Learning Agility, Progress Monitoring) and five personality (Openness to New, Risk Appetite, Result Focus, Agreeableness, Conscientiousness).

What a long session buys, and what it does not. Dynamic simulations with delayed consequences measure complex problem solving — a construct that overlaps with general cognitive ability by only around 18 per cent of variance (r ≈ .43 in a meta-analysis of 47 studies). It therefore contains something classic cognitive tests do not capture. A long session also allows behaviour to be observed both fresh and fatigued: the decline of sustained attention with time on task is one of the most replicable effects in cognitive psychology at group level. What we do not claim: that a long simulation yields higher predictive validity. The evidence points the other way — on the Sackett, Zhang, Berry & Lievens (2022) re-estimation, structured interviews come in at .42, job knowledge tests at .40, work samples at .33, cognitive ability tests at .31 and assessment centres at .29. We are not above that list; we are somewhere else on it.

We are equally careful about the game format itself. A systematic review of 34 studies (Ramos-Villagrasa et al., 2022) concludes that gamified formats do not offer advantages sufficient to recommend them over conventional methods — with the exception of improved candidate reactions. Our honest formulation is: comparable validity with substantially better candidate reactions. A randomised controlled experiment (Ohlms, Melchers, Kanning & Barends, Journal of Business and Psychology, 2025; N = 171) shows the game format reduces, but does not eliminate, deliberate faking — the authors put it as “GBAs are not a panacea to completely prevent it”.

The substantive difference lies in what a result is compared against. Requirements for the same role depend on where the company is in its life: computed across the 2,296 records of the handbook, 99 per cent of the 287 professions have requirement ranges at “Infancy” and “Bureaucracy” overlapping by less than half (median range overlap 0.40). Of 153 non-overlaps between the extreme stages, 152 fall on risk appetite; the swing of the range midpoint across the lifecycle is 18.8 percentiles for risk, 15.4 for openness, 11.5 for conscientiousness and 9.0 for cognition. Result focus barely moves. “Fit”, in other words, is a function of where the company stands, not a constant.

And a difference in explainability. Each of the 2,296 cells carries a human-readable rationale: 2,296 unique texts, 1.07 million characters, 464 characters per cell on average, with 32 per cent of them naming the rule that fired. In 303 cells (13 per cent) the benchmark permits high risk appetite alongside a low floor on cognition or monitoring — the method can say “fits on every parameter and still warrants attention”, which a single aggregate fit percentage cannot do in principle. We issue no hire/no-hire verdicts: this is ranking and role-fit assessment, and the decision belongs to the employer.

What the format difference changes in practice

The first practical consequence is task recognisability. A discrete, named module is easy to identify and dissect: preparation material for the Challenge tests is published both by the vendor itself — Aon's official page is called Prepare for your Online Assessment, and it remains our principal source for the durations — and by third-party prep vendors such as JobTestPrep and GraduatesFirst, from whom we took our second, independent confirmation of those durations. This is a property of the format rather than a charge against any vendor: a discrete task with a name of its own tends to accumulate walkthroughs over time. Whether and how preparation moves scores is an empirical question, and we have no data on Aon's products.

The second is modularity versus a single pass. A battery lets an employer assemble exactly the constructs it needs and stop at six minutes if the job is fast early screening. A continuous simulation offers no such flexibility: there is no way for it to become a six-minute instrument. That is neither a virtue nor a flaw but a constraint — and in the next section we name outright the situations where it settles the question against us.

The third is what ends up in the report. A battery of independent modules naturally yields a set of separate scores on separate constructs. A single-pass simulation yields a sequence of decisions in one context, where early moves change the conditions of later ones. These are different kinds of evidence and they do not convert into one another: ranking them “by accuracy” without a shared sample and a shared criterion would be a stretch, and we do not attempt it.

The fourth is compliance. Under New York's rules the law triggers not on the presence of AI but on the weight the output carries in the decision about a candidate. The EU AI Act's prohibition on emotion inference (Art. 5(1)(f)) is framed around inferences drawn from biometric data, and Illinois BIPA around the collection of biometric identifiers; how those provisions apply to an assessment running on behavioural telemetry — clicks, timings, choices — is a question for the client's counsel, and we offer no conclusion on it here. Camera proctoring, face recognition or eye-tracking change that picture in both directions. This is a property of the format, not a characteristic of any particular vendor.

How we checked this, and what is deliberately absent

Every link in this material was opened and checked on 4 August 2026. Figures about Aon appear only with a link to their source in the list at the foot of the page; figures about NeuroFrame come from a single internal register of metrics rather than being typed into the copy by hand. Where a vendor source disagrees with a third-party one (digitChallenge: 6 minutes on Aon's page against 5 at JobTestPrep), we show both values rather than picking the convenient one.

What is deliberately absent. We do not describe another company's validity evidence as “sufficient” or “insufficient”: we have neither their data nor any mandate to make that call. We do not assert that Aon fails to publish anything — we report only what we did and did not manage to find in open sources in one pass. We do not retell complaints as established fact, and we do not reproduce lists of Aon clients sourced from preparation vendors.

A specific note on proceedings. As far as we can tell, the ACLU's FTC complaint is unresolved at the date of publication, and we found no publicly announced regulator decision. If one appears, this page must be updated — we re-check the complaint's status before every revision.

Trademarks. Aon, cut-e, smartPredict, ADEPT-15, gridChallenge, switchChallenge, motionChallenge, digitChallenge, gapChallenge, chatAssess and vidAssess-AI belong to their respective owners and are used here nominatively only, to identify the product under discussion. NeuroFrame is not affiliated with Aon plc and is neither its partner nor its reseller.

And, once more, the point we opened with: this page is written by an interested party. The only protection against bias we can offer is checkability. Every statement about another company's product carries a link to a primary source — to Aon's own publication where the claim is Aon's, and to the third party's publication where the claim is theirs. Our own weak points we name ourselves, in the next section.

Questions

Are cut-e and Aon Assessment the same thing?

They are the same product family under different names at different times. cut-e GmbH was founded in Hamburg in 2002; Aon plc announced its acquisition on 4 May 2017 with financial terms undisclosed. Aon's own materials phrase the merged brand as “cut-e and Aon, as Aon Assessment Solutions”. Today the products sit on aon.com under Talent and Rewards; when we requested assessment.aon.com on 4 August 2026 it returned a 301 redirect to that section.

How long does an Aon assessment take, and how long does yours?

With Aon, duration is counted per module: as of 4 August 2026 the official candidate page lists gridChallenge at 9 minutes, motionChallenge, digitChallenge, gapChallenge and switchChallenge at 6 minutes each, and chatAssess at 20 minutes; ADEPT-15 runs about 25–30 minutes per the ACLU model cards. On the open pages we checked we found no total battery time — the employer assembles the set. NeuroFrame is one continuous 30–60 minute session, and it cannot be split into short pieces.

What exactly has Aon published about the validity of its gamified tests?

The main open document we obtained is a vendor presentation at the GBA Workshop (August 2019). It reports gridChallenge–G.A.M.E. equivalence at r = .71 with N = 306 on an Amazon Mechanical Turk sample, no mobile/desktop difference (d = −.12, N = 350), candidate reaction data and an Adverse Impact section. We found no indication that these materials went through peer review. Beyond that, per the ACLU's description, Aon's gridChallenge technical documentation runs to more than 70 pages and contains reliability and validity sections — we were unable to open those documents during our check, so we do not characterise their contents.

I have heard about a bias complaint. What do the documents say?

In late May 2024 the American Civil Liberties Union filed a complaint with the US Federal Trade Commission, “In the Matter of Aon Consulting, Inc.”, under Section 5 of the FTC Act, concerning ADEPT-15, vidAssess-AI and gridChallenge. The filer cites Aon's technical documentation and reports score differences between groups in the study samples. We found no publicly announced FTC decision as of 4 August 2026. This is a filed complaint — the filer's position, not an established fact and not a regulator's finding; we offer no assessment of the merits and would not advise drawing conclusions from it.

Which of you has an independent bias audit?

We found no such document publicly available for either of us, and that needs to be read correctly. Under New York's Local Law 144 rules, the duty to publish an annual independent audit falls on the employer or employment agency, not on the tool's supplier; those rules do not require the vendor itself to publish one. NeuroFrame has no formal bias audit and no adverse impact ratio report today, and we say so plainly. For Aon's products we did not find a separately published audit in open sources — that is the result of our search, not a claim that none exists.

We need assessment in Arabic. Who fits?

NeuroFrame has no Arabic version — neither the instrument nor the profile handbook — and that is the end of our answer. Aon publishes the phrase “more than 40 languages”, but on the open pages we checked we found no list of the specific languages and no information on which of them cover the gamified modules in particular. The right way to settle it is to ask the vendor directly, requesting confirmation for the specific modules rather than the line as a whole.

Why a 30–60 minute session when a competitor fits in six?

Because we measure something different, not the same thing for longer. Dynamic simulations with delayed consequences measure complex problem solving — a construct overlapping general cognitive ability by roughly 18 per cent of variance (r ≈ .43, meta-analysis of 47 studies). A long session also lets behaviour be observed both fresh and fatigued. What we do not claim: that a long simulation delivers higher predictive validity — on the Sackett et al. (2022) re-estimation the structured interview leads at .42, ahead of simulations. And if your job is fast early screening, six minutes beats us without qualification.