Advai supplies AI testing and monitoring, with MOD evaluation work and the FCA's live-testing programme providing concrete evidence of its role.
Advai sells the ability to evaluate and monitor AI systems. Its role is relevant to organisations that have already encountered promising models and now need evidence about which one to use, whether it is ready for a particular task and how its behaviour changes after deployment. That puts the company in a different part of the market from a supplier whose principal product is the model or end-user application itself.
The London company has publicly documented work with both the Ministry of Defence and the Financial Conduct Authority. Those relationships make the profile more than an account of a testing platform's advertised features. They show how evaluation expertise can become part of a public organisation's process for adopting technology.
Companies House records Advai Ltd, number 12540124, as a private limited company incorporated on 31 March 2020. Its significant-control register names David Michael Sully and Christopher George William Jefferson, each with more than 25% but not more than 50% of shares.
The disclosed bands establish meaningful ownership interests without supplying exact percentages. Advai's FCA collaboration announcement identifies Sully as chief executive and co-founder. The company is therefore identifiable through both statutory records and its public commercial work.
On its About page, Advai describes testing and monitoring as its sole business and says it does not sell AI solutions that compete with those being evaluated. That is the company's explanation of its market position. It is useful when assessing the relationship, while remaining distinct from an independently issued guarantee of impartiality in every engagement.
The Advai platform combines model selection, testing before deployment and monitoring afterwards. Model Arena addresses comparison between models, suppliers and configurations. Insights concerns test plans, traceable results and records for review. Monitor assesses system logs for performance, risk and security indicators.
These components correspond to different customer decisions. Selecting a model answers which candidate should progress. Testing it in context answers whether a proposed use is ready. Monitoring concerns what happens once the system is operating and its inputs, software or users change.
Advai also advertises managed setup for defining use cases, thresholds and initial evaluation work. This makes the offering a combination of software and expertise rather than a promise that buying a licence will settle every application-specific question.
The commercial value depends on how the evaluation fits into a real decision. A useful output might be comparable evidence for choosing a supplier, a record supporting an internal release decision or an alert that prompts a review. Those are different deliverables, and the customer should be able to identify which one it is commissioning.
A DASA case study published on 1 April 2025 traces support for Advai back to 2020. It describes several funded projects concerned with AI assurance and later work involving defence evaluation.
The history matters because Advai's commercial proposition did not begin simply with the recent popularity of conversational AI. The company has been building a business around understanding system limitations, then applying that expertise to different types of model and customer problem.
The government case study establishes programme support and a development history. Its positive assessment is still a published account by a supporting organisation, not a consolidated measure of revenue or an assurance result for every current product. The useful business evidence is the continuity of the evaluation role across successive projects.
On 10 November 2025, the Ministry of Defence described the AI Model Arena, developed with Advai and the National Security Strategic Investment Fund. The intended purpose was to compare supplier models against defence use cases and help identify candidates for further work.
The statement connects the project to earlier Royal Navy and Accelerated Capability Environment activity. It also distinguishes initial assessment from the more comprehensive evaluation associated with deployment. This makes the buyer's purpose clear: the platform supports a selection process rather than automatically authorising the use of every successful model.
The same release targeted a live platform by March 2026. That historical target is not sufficient to establish a current intake window or today's available challenges. For this company profile, the verified fact is Advai's named development role and the buyer's stated purpose.
In its November 2025 explanation, Advai describes shaping evaluation around customer needs and comparing submitted models under common conditions. The analytical implication is that the company can support the purchasing organisation as well as the developer being assessed.
That position gives Advai a potential relationship with the process of buying AI. The breadth of models evaluated should not, however, be confused with a count of software vendors purchasing Advai licences themselves.
The FCA's December 2025 announcement identifies Advai as its technical partner for AI Live Testing. The first group included firms such as NatWest, Monzo and Santander. The programme concerned AI applications being considered for use in financial markets, with support covering evaluation and monitoring questions.
This demonstrates a route beyond defence. The regulator supplies one part of the programme, Advai supplies technical support, and participating firms bring their applications. The named financial institutions are programme participants; the announcement does not establish that each signed a separate direct commercial contract with Advai.
On 21 April 2026, the FCA announced a second cohort, again naming Advai as technical partner. Eight firms were selected, including Barclays, Experian, Lloyds Banking Group's Scottish Widows and UBS. The stated applications ranged across customer-facing and business uses.
The repeat cohort provides stronger evidence of an ongoing programme role than a single partnership launch. The FCA scheduled testing through the end of 2026 and an evaluation report in the first quarter of 2027. Those later outputs remain relevant follow-up evidence; participation itself is not a blanket regulatory approval of the models being assessed.
The University of Edinburgh's EPCC announcement of 1 May 2024 identifies Advai as an early commercial user of its AI infrastructure. The relationship supports research into how AI systems behave and fail, with Advai providing feedback to the infrastructure team.
This is a different commercial position again: Advai is using a research facility's computing resources, rather than receiving the facility's entire investment budget as company funding. EPCC's wider infrastructure expenditure should therefore remain outside any calculation of Advai's revenue or capital raised.
The filed accounts for the year ended 31 March 2025 report an average of 21 employees, compared with 15 in the preceding year. These unaudited accounts omit the income statement. The directory consequently records the historical workforce average and leaves turnover unfilled, rather than deriving sales from balance-sheet balances.
The Second Front Systems profile concerns another part of bringing software into demanding customer environments. The Vizgard profile describes application software that can sit on the other side of an evaluation relationship. These comparisons help distinguish the supplier of an application from the businesses supporting its deployment and assessment.
Advai's defining commercial evidence is the combination of a testing platform, named public-sector programmes and continuing work in a second regulated sector. The next useful records will show how those programmes develop and how evaluation work translates into recurring monitoring relationships, with each customer's role kept clear.
Public company, government, FCA and EPCC text was readable. Filed iXBRL accounts were read through ordinary unauthenticated HTTP after the browser encountered a stale download redirect. They report 21 average employees for FY2025 and omit the income statement. The MOD's November 2025 March-2026 launch target is not treated as proof of current Model Arena intake. FCA cohort participants are not asserted to be separate direct paying Advai customers.
Suggest a correction