FE fundinfo

Nearly the entire business would run without the models. Fund data mastering, dissemination and feeds would remain untouched, along with regulatory document and template production, fund research and cashflow planning. So would pension switching, managed portfolio data, ratings and the investor facing site. The company itself describes Nexus AI as an enablement layer inside an existing platform rather than a product in its own right.

The models still matter because they sit inside regulated workflows. Document inspection determines which disclosures are flagged as non compliant against a named rulebook, and extracted meeting data is written into the fact find that underpins a suitability assessment.

There is a named control with a clear position in the execution path, described as an interaction rather than asserted as a principle. Where other transcription tools write extracted data onward, this product identifies where records need updating and prompts the adviser to review the proposed changes before they are mapped into the fact find. A human confirms each write into the planning file. Document inspection likewise runs ahead of human review rather than replacing it.

The limit is scope. The review gate is described for the data capture step only, with nothing equivalent stated for generated recaps, reports or conversational answers. Nothing sets out escalation rules, configurable thresholds, or what happens when an adviser rejects a proposed change.

The product is described as explainable and compliant by design and as delivering relevant, accurate answers, and none of that is measured. No error rate is published for document inspection, for data extraction from meeting transcripts, or for generated client documents, and no evaluation methodology, test set or independent assessment exists.

The count of more than six hundred rules checked describes coverage, not accuracy, and the two invite conflation: a system can test against every rule and still misclassify. For a component that flags regulatory documents as compliant or not, and that writes client facts into the file underpinning a suitability decision, the central gap is the absence of any published measure of how often it is wrong in either direction.

The company maintains a client success stories library and is long established infrastructure in its market, which places it well above a vendor with nothing to show. For the AI line specifically, the evidence is thinner than the company's stature would suggest.

Every figure attached to Nexus AI is vendor stated: thirty to sixty minutes saved per meeting, up to fifty percent less administration, and a research finding that most advisers spend a fifth of each meeting collecting data, all produced by the company itself. No named adviser firm or asset manager is attached to the AI product, and no independent party with anything at stake has assessed it. A direct competitor wrote the one published third party evaluation, a comparison.

The stewardship half is answered plainly and without hedging. The company states that no client data is used to train third party models, that no data is shared externally or used to train public models, and that all processing happens within its own infrastructure. It adds that the system is built and maintained by its own product and engineering teams.

For a platform recording client meetings and holding the financial circumstances of retail investors, that is the boundary statement that counts, and it often goes unstated in this market. The safety half falls short. Nothing describes evaluation of output quality, controls against fabricated content in documents that become regulatory evidence, or how model behavior is monitored once deployed. The compliance checking component was described as still in development after the launch.

The documentation set is real and reachable. It includes a published privacy policy, a cookie notice, website terms and a stated position of full compliance with European data protection law, applied consistently across the AI line and the established products. The commitment that client data is not used to train external or public models is the most material privacy statement for this product type, and it is made without qualification.

No privacy management certification is held, and no retention or deletion period is published for meeting recordings, transcripts or extracted client data. No subprocessor list exists, and nothing addresses how personal data moves between the group's United Kingdom, European and Asia Pacific operations.

The company's own legal and policies page lists ten documents. They are a privacy policy, a prudential disclosure, an investment firm disclaimer, a modern slavery statement and an environmental policy. The rest are an order execution policy, corporate responsibility material, website terms, a whistleblowing policy and a stewardship approach, alongside a gender pay gap report. Not one is a security certification, attestation or scope statement, and there is no trust center.

The security claim in the product material is the phrase secure infrastructure, with no external validation. A group that publishes a prudential capital disclosure and a gender pay gap report is plainly willing to disclose where it sees a reason to. The silence on information security is a choice about emphasis rather than an inability to document it.

Part of the group is itself authorized, which gives it more regulatory standing than a typical technology supplier. An investment firm subsidiary publishes both a prudential disclosure under the domestic investment firm regime and an order execution policy, documents only an authorized firm produces. Separately, the product line is built against named regimes rather than generic compliance language.

Document inspection tests against more than six hundred consumer composite investment rules ahead of the regulator's 2027 deadline. The wider production line covers packaged retail investment, collective investment, sustainability and climate disclosure regimes across jurisdictions. The authorization attaches to the investment management subsidiary rather than to the AI line, and no regulator has assessed, tested or supervised the models themselves.

Nothing published addresses whether the system performs evenly across the people it processes. The exposure is concrete for this product. Transcription accuracy varies measurably across accents and speech patterns, and this platform turns recorded conversation into the structured client facts that populate a fact find and inform a suitability assessment. A transcription failure becomes a data quality failure in a regulated client file.

No demographic analysis, evaluation of transcription performance across speaker populations, bias testing methodology or AI governance framework is published. The company publishes environmental, modern slavery, stewardship and pay gap disclosures, so the absence of any equivalent statement covering model fairness is conspicuous.

No accuracy guarantee, remediation commitment or allocation of responsibility between the company and the firms relying on its outputs is published. Two distinct affected parties have no stated position. The adviser firm acting on an incorrectly extracted client fact, or on a document wrongly passed by automated inspection, has no published recourse against the supplier. The retail client is further removed still.

Their recorded conversation becomes structured data underpinning a recommendation, and they have no route to establish how their circumstances were captured or interpreted if advice is later disputed. The review prompt before data is written to a fact find is a real control, but it places the burden on the adviser rather than creating any obligation on the vendor.

The company acknowledges third party models in its stack, since it commits that client data will not be used to train them, and then names none of them. No base model, provider, version or component is identified for transcription, extraction, generation, document inspection or the conversational interface, and no subprocessor list is published. The framing points the other way, emphasizing that the system is built and maintained by the company's own teams within its own infrastructure. A build claim stands in for a disclosure. At least one far smaller rival publishes a dated table naming every model provider, its processing activity and its country of processing.

The integration position is structural rather than negotiated, because the company owns the systems it integrates with. Nexus AI runs inside FE CashCalc and FE Analytics, the planning and research tools a large part of the United Kingdom advice market already uses daily, and writes extracted client data directly into the CashCalc fact find rather than exporting it.

Independent review of the product identifies this as its principal advantage, noting low implementation friction where a firm already runs the stack and that data stays inside the existing environment, and describes read and write connectivity to around a dozen United Kingdom back office systems. On the institutional side the group's dissemination infrastructure already moves fund data and documents to platforms, distributors and regulators. The one qualification is that the specific back office count comes from third party review rather than the vendor's own material.

Delivery is hosted software accessed through the company's platforms and portals. The residency position is an assurance about environment rather than a statement of location. Processing is said to happen within the company's own secure infrastructure, and independent review describes data as staying inside that environment, but no hosting region, country or in country residency option is named. That gap is material for this group specifically.

It operates across the United Kingdom, continental Europe and Asia Pacific with subsidiaries in Germany and Australia, so a buyer cannot establish which jurisdiction will hold recordings of its client conversations. No data transfer mechanism or subprocessor footprint is set out.

Every route to a price resolves to a contact form, a demo request or a sales conversation. No rates, tiers, bundles, seat pricing or trial terms are published for Nexus AI or for the underlying platform products. The AI layer is presented as embedded within an existing subscription, and nothing published says whether it carries an incremental charge. For a product explicitly sold on low friction adoption by firms already inside the stack, that omission is notable, since incremental cost is the first question an existing customer would ask.

Both sides of the fund distribution chain are served, from the asset managers who create products to the advisers who select them. Segment coverage is broad and evidenced by distinct product lines rather than landing pages. The company addresses investment managers, financial adviser and paraplanning firms, and institutional wealth managers.

It also serves private and retail banks, platforms and technology providers, along with data partners and management companies, each with its own product set and stated problems. Geography spans the United Kingdom, continental Europe and Asia Pacific, carried by group brands including a German structured products research business and two Australian research and ratings houses. The regulatory production line covers multiple jurisdictional regimes simultaneously.