Intelliflo

Intelliflo office has been the back office of record for United Kingdom advice firms for around two decades. It holds client data, the fact find, workflow, onboarding and reporting, and inference arrived in February 2026 as a suite layered on top of it. Strip the models out and the platform runs exactly as it did.

The shipped module still does real work rather than decorating: converting handwritten notes, documents and audio into structured updates across more than a hundred and ninety fields is throughput a person would otherwise carry. The shipped and the announced are separate matters, because three of the four modules described are not yet live.

The gate is explicit and universal, stated in the vendor's own material and in independent coverage of it: every generated update requires adviser review and approval before it is finalized. It is expressed without qualification, and it sits at the point where it matters, between extraction and the client record. The forthcoming advice module carries the same design in a harder place.

The adviser chooses whether to act on a suggested recommendation or take a different course and is described as in control throughout, and the company states that audit trails are preserved rather than bypassed. One tension lies ahead. The announced next generation platform is to run agentic capability continuously in the background, and a design that scans and flags is easier to reconcile with a universal approval gate than one that acts.

The extraction claim is precise about scope and silent about correctness. More than one hundred and ninety fact find fields are populated from handwritten notes, documents and audio, and the records produced are described as precise and compliant. No accuracy rate, error rate by input type, benchmark, validation methodology, drift statement or revalidation cadence is published for any of it.

Handwriting and audio are the two input types where extraction error is most likely and least visible, and an incorrect income figure or misheard objective entering a fact find becomes the basis for a suitability assessment. The approval gate is a real mitigation and it depends on an adviser noticing, which is a different thing from a measured error rate.

Neither of the two quantified claims in circulation is attached to an identified firm. One says administrative time can fall by as much as eighty five percent, and the other that advisers using the platform more than doubled weekly client onboarding. A test phase with early adopters is referenced without naming any of them. The one named individual in all the material is the company's own chief executive.

None of the published material names a customer, offers a case study or cites an independent evaluation. A platform of this size in a market this concentrated will almost certainly have customer stories, and they are not reflected here.

Nothing published says whether client meeting audio, fact find content or the documents advisers upload are used to train, tune or evaluate any model. Nor does it say whether content is retained by any model provider, or whether material from one advice firm can inform output shown to another. The concentration raises the stakes.

A platform holding the client conversations of a substantial share of one national advice market accumulates a body of data with no equivalent elsewhere in that market, and nothing published states what may or may not be done with it. A direct denial would be a strong disclosure, and a competitor in the same market publishes one.

A fact find carries income, health, family circumstances, vulnerability indicators and objectives for individual consumers. The shipped module now captures client meetings as audio and converts them into more than a hundred and ninety structured fields. No retention schedule for meeting audio, deletion commitment or statement of how long recordings persist once converted is published. Whether the recording is discarded after extraction or retained is the question a firm's own data protection assessment would open with.

Product and announcement material names no attestation, information security standard or audit report, and no dedicated security or trust page is reflected here. The platform holds the client records of a large share of a national advice market, which makes the gap consequential. It also makes a published position likely to exist somewhere, most plausibly a trust or security page carrying a numbered standard and an audit report type.

A technology supplier to advice firms holds no license and requires none. The company builds close to the regulatory perimeter without standing inside it. Announced modules are described as reviewing suitability reports and fact finds against a named regulator's requirements and supporting a senior manager accountability regime and regulatory reporting. That is a product capability aimed at helping firms meet obligations rather than any supervised status of its own. The firm carries the obligation and the vendor carries none of it, and nothing published addresses how a firm evidences to an examiner which parts of a suitability file were machine assessed.

Two exposures sit here and neither is addressed. Audio extraction performance across accents, dialects and speech patterns determines the quality of the client file that results. Vulnerability indicators are among the fields being captured, which is precisely the population where a missed cue carries regulatory and human consequence.

The announced advice and quality modules raise the second: a rules based engine that suggests recommendations and an automated review that flags conduct risk both make distinctions across clients and advisers at scale. No evaluation across speaker populations is published, and nothing explains how suggestions and flags are derived or whether they were examined for systematic effects.

No warranty, accuracy commitment, service level, correction obligation or allocation of responsibility between vendor and advice firm is published, and neither is a disclaimer placing that responsibility plainly on the firm. The approval requirement carries an unstated implication. By requiring the adviser to approve every generated update, the design places the professional squarely behind the output, which is sound practice and also a transfer of exposure. A firm whose suitability assessment rested on a misextracted fact find field has approved it, and nothing published addresses what recourse exists.

A development partner is named, but the model behind the shipped product is not. The suite is stated to have been developed in collaboration with a named wealthtech AI partner. That tells a buyer who helped build the capability and, for the announced advice module, whose algorithmic engine sits behind the recommendations.

What remains unidentified is the layer that does the work in the shipped product: no provider, model family, version or inference host is named for the imaging and audio processing that reads a client's handwriting and listens to a client's meeting.

Intelliflo is the system others integrate into rather than the one seeking connections, and its stated direction is to lean into that with a platform it describes as open. The AI design reflects this. Rather than shipping a standalone assistant, the company embeds the capability in the existing adviser workflow, on the reasoning that data fragmentation across multiple disconnected systems is the underlying problem in advice firms. Inbound capture covers handwritten notes, PDFs and audio. There is no public connector catalog, and no platform, provider, custodian or planning tool is named individually. For a system of record those specific names are what a firm checks first.

The platform is described as cloud based, and nothing further about topology is published: no hosting region, storage location, tenancy statement or residency commitment. The omission matters less because the customer base and the regulatory frame are both single jurisdiction, but it is not answered. A firm cannot establish from published material whether its client records and captured meeting audio remain within that jurisdiction.

No rate, per adviser or per seat figure, or tier structure is published, and nothing says whether the AI suite is included in an existing subscription or charged separately. That last question is the consequential one. Firms already pay for the underlying platform, and whether the new suite arrives inside that fee or alongside it determines whether adoption is a decision or a default.

Depth in one market rather than reach across many. The platform serves financial advice firms across the United Kingdom at every size, from a single adviser practice to a national network. Its position is substantial enough that the company's own adviser research is quoted as market data by the trade press, a reasonable proxy for penetration.

The buyer type is narrow by design: regulated advice firms and the paraplanners and administrators inside them, rather than banks, insurers or asset managers. Geography is the harder limit. The published material describes a single jurisdiction and a product built around one regulator's requirements, which is a strength commercially and a limit on reach.