AdvisoryAI

The removal test leaves templates and connectors. Models transcribe adviser client meetings with speaker identification, extract financial position, income, expenditure, assets, liabilities, existing provisions and protection needs from unstructured conversation, recognise stated and implied objectives such as retirement, education funding and inheritance planning, synthesise attitude to risk and capacity for loss discussions into a risk category, draft the suitability report into the firm's own template, and answer questions across the client book with citations to the record.

Every one of those is model work and none of it survives the models being taken out. The compliance check layer that sits on top is the only rules based component and it exists to verify model output.

The boundary is stated and then justified, which is rarer than stating it. The company publishes that the machine drafts the report and the adviser and paraplanner review and approve it, and adds that this is not a disclaimer but a reflection of how regulated advice actually works, then explains why by reference to the specific conduct rule requiring the report to set out the client's demands and needs, explain why the recommendation is suitable given their objectives, situation, knowledge and experience, and describe the disadvantages of the transaction, reasoning the company says requires professional judgement.

That is an argument rather than a policy line. The compliance layer reinforces it by catching gaps before a report reaches the compliance team, so the machine prepares the reviewer's work rather than replacing the reviewer. A different shape from Recordsure, which prohibits the decision architecturally, where this justifies the prohibition from the rulebook.

The verification layer is unusually concrete for this index. Forty two automated checks run on every suitability report plus multi category checks on fact finds, and the output is not a score but a structured report showing pass or fail per category, a percentage, and specific remediation guidance for each failed item, which is a machine checking machine output in a form a compliance officer can act on. Book wide answers are stated to cite the record, so an assertion can be traced.

Independent trade reporting indicates machine generated reports require around 40 percent fewer compliance corrections than manual ones, which is an accuracy adjacent comparison from outside the vendor. What is absent is extraction accuracy, the number that matters most, since every check downstream assumes the facts were heard correctly in the first place.

More than 2,000 advisers across more than 400 United Kingdom firms, with a substantial listed wealth manager named among them. Independent corroboration is unusually direct: a trade title ranked the platform the most used artificial intelligence system among United Kingdom advisers in the first half of 2025 and rates it first in its category, which is a market position measured by someone other than the vendor.

Outcomes are quantified against the tasks advisers actually track, with meeting notes falling from an hour and a half to fifteen minutes, suitability reports from four to six hours to under one, and letter of authority pack summarisation from three hours to one for ten packs. The strongest claim is a business rather than a time metric: firms are stated to have doubled client capacity per adviser within a year, with one adviser onboarding nineteen new clients in their best year.

This vendor answers the cross customer question in the benchmark wording and then improves on it. The statement that models are never trained on client data is the Rulebase formulation, which the index has treated as the standard for foreclosing the boundary in one line. AdvisoryAI goes one step further by naming what is used instead, with anonymised material applied to tone of voice and template training only and configuration changes contained to the firm that made them.

Disclosing the narrow exception is more useful than a blanket denial, because a buyer can evaluate whether that specific use is acceptable rather than wondering what the denial excludes. Reports are generated into the firm's own templates rather than a vendor house style, which keeps the firm's work product its own.

Four commitments stack and they are stated plainly rather than implied. Client data is never used to train the models. Data residency is specified, with hosting in United Kingdom data centres on United Kingdom based infrastructure. Encryption is stated both in transit and at rest. Compliance with the European data protection regime is named alongside an information security certification.

What lifts this to the top grade rather than leaving it at strong is the precision of the exception: anonymised data is used for tone of voice and template training only, and configuration changes stay within the firm, so the vendor discloses the narrow thing it does use rather than making a blanket claim a buyer would have to take on trust. Third privacy A in the index after Zeplyn and Rowspace.

Two named frameworks appear, a government backed baseline cyber certification and the international information security standard, alongside stated encryption in transit and at rest. One inconsistency belongs on the record: current material states the information security certification is held, while earlier published material describes it as in progress, so a buyer should ask for the certificate and its scope rather than relying on either statement. Two named frameworks with a verifiable certificate would place this among the strongest positions in the index; the ambiguity is what holds it at B rather than higher.

The most precise regulatory citation recorded in this index. The company cites the specific conduct rule provision governing suitability reports by its exact reference and sets out what that provision requires, rather than naming a regulator or a regime in the abstract, and it uses that requirement to explain a product design decision.

Beyond it, the broader conduct sourcebook, the consumer outcomes regime with its four areas explained, the senior manager accountability framework and the regulator's expectations all appear in published material, and the compliance checks are mapped to named obligation categories including anti money laundering documentation, client profiling, risk assessment adequacy and recommendation justification. Seventh A on this axis, and the only one where a vendor cites a rule at provision level.

The speech recognition variance finding applies for the ninth time in this index and lands differently here because the subject is the client rather than the employee. Extraction works from natural conversation in adviser meetings, and accuracy varies with accent, dialect, age, hearing difficulty and speech impairment, so a client whose speech the system reads less well may have their circumstances, objectives or risk tolerance captured less accurately, and that misreading flows into a document recommending what they should buy.

Risk profile extraction is the sharpest instance, since synthesising attitude to risk and capacity for loss from conversation is an inference about a person that shapes what they are sold. The mitigation is genuine, since an adviser reviews and approves and 42 checks flag gaps, but no per accent or per demographic extraction accuracy is published and no analysis of performance across client populations was located.

Accountability is allocated clearly and the allocation is correct rather than convenient. Because the adviser and paraplanner review and approve every report, and because the conduct rules place responsibility for suitability on the regulated firm, there is always an identifiable accountable person behind the document a client receives, and the compliance layer gives the firm a mechanism to find defects before that happens. Book wide answers citing the record make assertions checkable.

The client is not left without route either, since a report their adviser approved sits inside the ordinary complaints and ombudsman path regardless of how it was drafted. What is missing is anything binding the vendor itself, and any correction mechanism for a client whose circumstances were extracted wrongly and who has no way to know it.

The infrastructure provider is named and the back office platforms in the data path are named, so part of the chain is visible. The model layer is not. This is a generative product drafting regulated documents from meeting audio, which implies external model providers somewhere in the stack, and no provider, hosting arrangement for inference or subprocessor list was located.

The training claim states models were built on thousands of sample reports by a team of former advisers and paraplanners, which describes the tuning corpus without identifying the base model underneath it. Given how precisely this vendor discloses its data handling, the silence on the model layer is the one conspicuous omission in an otherwise unusually transparent profile.

Native integration covers effectively the whole back office market this segment runs on, with four named adviser practice management platforms connected directly and meeting notes, reports and client data syncing back automatically rather than being exported. Capture spans the three major video conferencing platforms plus a mobile application for in person meetings, so the product works wherever advice actually happens.

Reports populate the firm's existing templates rather than imposing a house format, which removes the change management objection that usually blocks adoption. Covering the entire connector surface a buyer could plausibly need, in a market where four platforms hold the field, is what this grade is for.

Residency is stated rather than left to inference, with hosting in United Kingdom data centres and client data held on United Kingdom based infrastructure from a named cloud provider. That is more than most vendors in this index disclose and it matters for a buyer whose own regulator expects to know where client files sit.

Held below the top grade because the commitment is a statement of current arrangement rather than a contractual position a customer can set, no region selection or private deployment option is offered, and nothing describes what happens to residency if the company expands beyond its home market.

No pricing, packaging or basis of charge is published and every route in is a demo request. Onboarding effort is stated concretely, with standard templates usable from day one and custom template configuration completed within two weeks by the company's own ex paraplanner team, which tells a buyer what adoption costs in time if not in money. Nothing indicates whether charging runs per adviser, per firm, per report or per seat, which matters for a segment where firm sizes range from a single adviser to hundreds.

Coverage is deliberately narrow and the narrowness is the strategy rather than a gap in execution. This serves United Kingdom financial advice firms and nothing else, with no banking, lending, insurance or institutional business, and no evidenced footprint outside one jurisdiction. Within that segment it reaches the roles that matter, covering advisers, paraplanners, compliance staff and back office, and the depth is real.

But the axis measures breadth of institution and segment served, and one firm type in one country is the position, which is what separates this from Zeplyn and Nevis serving multiple wealth structures across markets.