Nitrogen

Fifteen years of history make the order plain: the platform came first and the inference arrived on top. 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 keep working without it.

The quantitative scoring at the heart of the platform is financial mathematics rather than an inference capability, so the Risk Number does not count as a model. Real inference now sits on top, in the agentic engine launched February 2026, in statement conversion and in the tax product, and it does real work rather than decorating. It is also three quarters of a year old on a platform whose core was complete without it.

Oversight rests on two independent supports. The first is the gate. The agentic engine can be tasked with work that 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 further constraint narrows the blast radius, since the engine operates only inside the platform's own structured data rather than roaming open data.

The second support is external. The company states ISO 42001 certification, which is externally audited coverage of exactly the governance, oversight and lifecycle processes at issue, rather than a design intention published on a product page.

The scoring methodology is well provenanced, and the performance of the inference is unmeasured. The risk score is patented, and its theoretical basis in the behavioral economics of decision making under uncertainty is public and attributable. No accuracy figure on statement conversion, evaluation of the agent's output, benchmark, validation method or revalidation cadence is published. One disclosure comes by implication.

The vendor sells a human verified tier of its own statement conversion as a paid upgrade, a commercial acknowledgement that the unverified output requires checking, and one that 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.

No named firm reports a measured outcome, no efficiency or growth figure is attached to an identifiable customer, and there is no independent quantification of category share. A partnership announcement is a commercial judgment rather than a customer reporting what happened. The company has historically led the risk tolerance category of an independent adviser software survey, and a current measured share from that survey is not reflected here.

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 is not stated in either direction. ISO 42001 would require documented positions on training data provenance and on the lifecycle of data used in model development. That makes this a strong candidate for a disclosure that exists somewhere unpublished, most likely in the certification scope statement or a trust page.

No retention schedule, deletion commitment or processing terms for advice firms are published, and nothing explains how client financial data and uploaded statements are handled. The vendor's ISO 42001 certification would ordinarily require documented data governance. That makes a privacy policy and data processing terms more likely to exist unseen than to be missing.

Two certifications are named plainly: a SOC attestation and ISO 42001. The second is a governance standard for the development and operation of AI rather than a general information security one, and it is still uncommon among financial software vendors. The company also has a substantial shipped AI line for the standard to govern.

No dedicated trust page, penetration testing statement, continuous monitoring platform or scope for either certification is published, so the credentials rest on a press announcement rather than a maintained disclosure surface.

A software supplier holds no license and requires none. This product sits 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. Nothing published says what the firm is expected to 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.

Certified process, undisclosed results. ISO 42001 certification means an externally audited framework exists covering how models are developed, assessed for impact and monitored. What certification does not supply is any finding. No fairness evaluation is published, and neither is any testing of the risk questionnaire or the agent's output across client age, wealth level or demographic group, or any impact assessment.

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.

There is no published warranty, accuracy commitment, correction obligation or allocation of responsibility between vendor and advice firm. Two features allocate it implicitly and in the same direction. The approval gate before any action executes is a real 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.

The launch releases, product descriptions and trade coverage name no model provider, model family, version or inference host behind the agentic engine, the statement conversion or the tax analysis. The gap stands out against this vendor's other disclosures. A company certified to ISO 42001 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 requires, and nothing concerns whose model does the reasoning.

The outward reach is real, and it runs in both directions. Beyond connecting into other people's systems, this vendor packages its own scoring and portfolio analytics as a programmatic engine for institutions to embed. That is the deeper commitment of the two, because 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. No enumerated integration list names the CRM and planning systems an advice firm would need connected. For a vendor of this maturity an integrations directory is far more likely to exist than not, and it is not reflected here.

Hosting arrangements, storage region, residency commitment and tenancy are all unstated, and nothing says where uploaded statements and generated portfolios are processed. 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 published price exists, and it covers an add on rather than the platform. 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. Third party figures circulate for it, but they come from aggregators rather than the vendor. The disclosed price is for the smallest component, and the substantial spend cannot be quoted from published material.

Breadth by firm size is substantial and by institution type it is narrow, a shape common in adviser technology. 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. That is a different distribution route rather than a bigger version of the same one. The limit is that everything sits inside one jurisdiction's advice market, with no bank, insurer or asset manager segment addressed as a buyer in its own right.