Nitrogen

The established platform reading applies and the fifteen year history makes it unusually clear cut. The Risk Number, the portfolio alignment analytics, proposal generation, the investment research product and the stress tested income planning all predate any inference in this product by more than a decade and all survive its removal working.

There is a second reason beyond age, and it is the distinction this index has preserved since Prometeia: the quantitative scoring at the heart of this platform is financial mathematics rather than an inference capability, so a reader should not count the Risk Number as a model in the sense this axis measures. Real inference now sits on top, in the agentic engine launched February 2026, in statement conversion and in the tax product, and it is genuine rather than decorative. It is also three quarters of a year old on a platform whose core was complete without it.

This is the strongest oversight position located in the pocket and it clears the bar on two independent legs. The first is the gate: the agentic engine can be tasked with work that genuinely acts on the world, including sending a risk questionnaire to a client, converting a statement into a portfolio and drafting a proposal, and adviser approval is stated as required before any action executes.

A second constraint narrows the blast radius further, since the engine operates only inside the platform's own structured data rather than roaming an open corpus. The second leg is what separates this from vendors that assert a principle: the company states certification against the international management system standard for artificial intelligence, which is externally audited coverage of exactly the governance, oversight and lifecycle processes this axis asks about, rather than a design intention published on a product page.

The methodology behind the scoring is unusually well provenanced and the performance of the inference is unmeasured, and only the second is what this axis asks. The risk score is patented and its theoretical basis in the behavioural economics of decision making under uncertainty is public and attributable, which is more than most quantitative scores in this index offer.

Against that, searched for an accuracy figure on statement conversion, any evaluation of the agent's output, a benchmark, a validation method or a revalidation cadence, and located none. One disclosure by implication is worth naming: the vendor sells a human verified tier of its own statement conversion as a paid upgrade, which is a commercial acknowledgement that the unverified output requires checking and which prices the error rate rather than publishing it.

Named institutional relationships are plentiful and none of them reports a result. A national financial services firm is named adding these products, an automated investing platform's adviser business is named in an integration, and adviser platform operators are named in partnerships, all announced in the vendor's own releases or in trade coverage of them. The annual user conference draws firms across the full size range and sells out, which is a real if indirect adoption signal.

What is missing is the thing that would carry the top grade: searched for a named firm reporting a measured outcome, any efficiency or growth figure attached to an identifiable customer, or any independent quantification of category share, and located none. A partnership announcement is a commercial judgement rather than a customer reporting what happened. Queued check: this vendor's placement in the independent adviser software survey's risk tolerance category, where it has historically led and where a measured share would move this row.

Recorded as an unverified absence. Searched for any statement on whether client portfolios, questionnaire responses, uploaded statements or generated proposals are retained, used to improve the service, or used to train, tune or evaluate any model, and located none in either direction.

The certified artificial intelligence management system this vendor holds would require documented positions on training data provenance and on the lifecycle of data used in model development, which makes this a strong candidate for a disclosure that exists somewhere unpublished. Queued check: the certification scope statement and any trust page.

Recorded as an unverified absence. Searched for a retention schedule, a deletion commitment, processing terms available to an advice firm, or any account of how client financial data and uploaded statements are handled, and located none in the material opened.

The certification the vendor holds for its artificial intelligence management system would ordinarily require documented data governance, which makes the absence of a published position more likely to be an unopened page than an empty one. Queued check: a privacy policy and any data processing terms.

Two certifications are named plainly and one of them is genuinely rare in this index. A service organisation control attestation is stated, and alongside it certification against the international management system standard for artificial intelligence, which is a governance standard for the development and operation of artificial intelligence rather than a general information security one.

Only one other vendor across more than four hundred and eighty records has been found holding it, and unlike that vendor this one has a substantial shipped artificial intelligence line for the standard to actually govern. Held below the top grade because searched for a dedicated trust page, a penetration testing statement, a continuous monitoring platform or the scope of either certification, and located none, so the credentials rest on a press announcement rather than a maintained disclosure surface. Queued check: a trust or security page.

A software supplier holds no licence and requires none, and no penalty attaches. The position is worth stating because this product sits unusually close to a specific regulated obligation without discharging any part of it: a risk tolerance score exists to evidence that a recommendation was suitable for the client who received it, and that evidence goes into a file an examiner may later read. The obligation stays wholly with the advice firm.

Searched for any statement about what the firm must retain, how a score's provenance is demonstrated, or how an examiner would distinguish a target set by an adviser from one proposed by the agent, and located none.

Certified process, undisclosed results, and that combination is why this sits above the index norm rather than at the top. Certification against the international management system standard for artificial intelligence means an externally audited framework exists covering how models are developed, assessed for impact and monitored, which is a stronger position than almost anything else in this index can show and is notably stronger than the governance software vendors who sell fairness testing and publish none about themselves.

What certification does not supply is any finding: searched for a fairness evaluation, any testing of the risk questionnaire or the agent's output across client age, wealth level or demographic group, or any published impact assessment, and located none. A risk tolerance score that systematically reads one group as more conservative than another would steer whole cohorts toward different portfolios, and nothing published addresses whether that was examined.

Searched for a warranty, an accuracy commitment, a correction obligation or any allocation of responsibility between vendor and advice firm, and located none. Two features allocate it implicitly and in the same direction. The approval gate before any action executes is a genuine control and simultaneously places every consequence on the adviser who approved.

The paid human verification tier goes further, because charging separately for a checked output is an explicit statement that responsibility for an unchecked one lies with the buyer who declined to pay. The client whose tolerance was scored and whose portfolio was aligned to that score has no described route to see how the score was reached or to contest it.

Searched the launch releases, the product descriptions and the trade coverage for a model provider, a model family, a version or an inference host behind the agentic engine, the statement conversion or the tax analysis, and located none.

The gap is conspicuous against this vendor's other disclosures rather than routine, because a company that publishes certification against an artificial intelligence management system standard has demonstrably documented its model estate somewhere for an auditor. Everything the announcements say about the engine concerns what data it may reach and what approval it must obtain, and nothing concerns whose model does the reasoning.

The outward reach is real and it takes an unusual form. Rather than only connecting into other people's systems, this vendor packages its own scoring and portfolio analytics as a programmatic engine for institutions to embed, which is the deeper commitment of the two because it means other firms build against it.

Named platform relationships exist with an automated investing platform's adviser business and with adviser platform operators, and statement ingestion accepts uploaded documents from outside the estate entirely. Held below the top grade because searched for an enumerated integration list naming the client relationship and planning systems an advice firm would need connected, and located none in the material opened, which for a vendor of this maturity is far more likely to be an unopened page than an absent capability. Queued check: the integrations directory.

Recorded as an unverified absence. Searched for hosting arrangements, a storage region, a residency commitment, a tenancy statement or any account of where uploaded statements and generated portfolios are processed, and located none.

One architectural fact is disclosed and it is adjacent rather than sufficient: the agentic engine is stated to operate within the platform's own structured data environment, which describes the boundary of what the engine can reach rather than where any of it physically sits.

A real published figure exists and it covers an add on rather than the platform, which is an unusual split worth recording precisely. The human verified tier of statement conversion is announced at a stated monthly rate in the vendor's own release, so a buyer knows exactly what that capability costs before speaking to anyone.

The platform itself is contact sales, and the third party figures circulating for it are not usable under this index's rule against grading commercial terms from aggregators. Held below the top grade because the disclosed price is for the smallest component and the substantial spend remains unquotable from published material. Queued check: the vendor's own pricing page.

Breadth by firm size is substantial and by institution type it is narrow, which is the standard shape for this pocket. The published customer range runs from solo practitioners through ensemble practices and mid market firms to enterprises carrying more than a thousand advisers, and broker dealer firms are addressed alongside registered advice firms.

A separate programmatic engine product extends reach further by letting institutions and larger wealth enterprises embed the scoring and analytics in systems this vendor does not own, which is a genuinely different distribution route rather than a bigger version of the same one. Held below the top grade because everything sits inside one jurisdiction's advice market, with no bank, insurer or asset manager segment addressed as a buyer in its own right.