Ammonite AI

Generation is the whole product. Planbot ingests fact finds, meeting transcripts, policy documents, valuations and fee schedules and writes the firm's own template from that unstructured material. Without the models, what remains is the blank template the firm already had, which is the starting position the product exists to improve on. There is no workflow engine, record system or rules layer underneath to fall back on.

The positioning is candid, and it is doing real work. A co founder states publicly in trade press that the model can make mistakes and that a human check is therefore required at the end of the process to confirm the output remains compliant and suitable. The founders repeatedly frame advisers and paraplanners as remaining firmly in charge of the report writing process. Design supports the claim.

The firm owns and edits the template, changes it without vendor involvement, and receives a draft to review and export rather than anything sent onward. None of this is a mechanism. No checkpoint is enforced in the product, and no confidence indication accompanies generated passages. Nothing marks which content was drawn from which source document, and nothing records that a review took place.

Errors are admitted but not measured. A co founder states in trade press that the model can make mistakes, and marketing separately describes the output as having impressive consistency and accuracy. Neither position is quantified. No error rate is published for template population or extraction, and no evaluation methodology, test set or independent assessment exists. Nothing describes how often a generated report requires substantive correction rather than editing. Without that figure, a firm is no better able to size the review burden it is taking on.

The evidence runs ahead of a February 2026 launch. One customer firm, Simpson Wood, is named publicly, and its head of client servicing is quoted by name describing faster turnaround and progress through an annual review backlog. The company states users have more than halved the time taken to produce a suitability report. Coverage is independent, spread across four separate United Kingdom advice trade publications rather than the vendor's own channels.

The founders' background as former Chartered advisers who built and sold an advice business is externally verifiable. Three limits remain. No figure is attached to the named customer, whose quotation is qualitative. The halving claim is vendor stated across unnamed users. A founder's suggestion that the tool could triple adviser capacity is speculation rather than a measured result.

Neither half is addressed publicly. Nothing states whether uploaded fact finds, meeting transcripts, valuations and policy documents are used to train or improve models, or whether any third party provider is permitted to train on that material. Two direct competitors publish an explicit position on exactly that question. Nothing says whether a firm can decline such use, or how long uploaded material and generated drafts are retained.

On the safety side, no control against fabricated content is described for a document that becomes regulatory evidence. No evaluation of output quality is offered, and nothing sets out what the system does when the uploaded material is incomplete or internally contradictory.

Privacy is the least documented area for a product that receives complete client files, including pension policy documents and full financial circumstances, by upload. No ISO 27701 certification is held, and no retention or deletion period is published for uploaded documents, transcripts or generated drafts. No position is stated on the lawful basis for processing or on data subject rights over material supplied by the adviser rather than the client. Nothing explains how a client would learn their file had been processed by a third party system.

No ISO 27001 certification or SOC report is published, and there is no trust center, security page or scope statement. The material a firm uploads is among the most sensitive it holds. It includes complete fact finds and pension policy documents, plus portfolio valuations and fee schedules, for named retail clients. The file upload model means that material is transferred to the vendor by design rather than accessed in place. Advice firms are routinely asked by their networks and compliance oversight providers to evidence supplier security, and nothing published would answer such a request.

Ammonite holds no financial license and no regulator has assessed the product, which is the expected posture for a supplier. The regulatory grounding comes from the artifact it produces. Suitability reports and annual reviews are the mandated file level record that a recommendation was appropriate for a client, and the tool is built around those documents specifically rather than around generic productivity.

The founders' background as former Chartered advisers who ran a regulated advice business gives the product design direct practitioner grounding in the relevant conduct rules. No authorization, registration, program participation or regulator engagement exists, and no external party has assessed whether outputs meet the standard the documents are held to.

Nothing published addresses whether the system performs evenly across the clients whose files it processes. The product ingests meeting transcripts alongside documents, which carries the transcription variation problem common to transcript based tools. It also generates the narrative rationale explaining why a recommendation suits a particular person, so uneven handling of how different clients describe their circumstances would surface directly in the regulated document. No demographic analysis, bias testing methodology or governance framework covering model fairness is published, and nothing evaluates performance across client or speaker populations.

The founders are clear that responsibility for the final document rests with the adviser or paraplanner, which is the correct regulatory position, stated plainly. That settles the vendor's exposure without establishing anyone's remedy. No accuracy guarantee, remediation commitment or allocation of liability between vendor and firm is published.

The retail client has no route to learn that a model processed their fact find and policy documents and drafted their suitability rationale, or to establish how their circumstances were interpreted if the advice is later disputed.

No base model, provider, version or component is named for document ingestion, extraction or template population, and no subprocessor list is published. The company describes combining adviser and paraplanner experience with cutting edge technology, which characterizes the team rather than the stack, and leaves open whether the underlying models are its own or licensed from a third party.

No identifier exists that a firm could record against a generated suitability report to establish later which system produced it, which matters for a document that may need to be defended years after it was written.

The product deliberately sits outside the firm's systems rather than inside them. It runs on a file upload workflow, where documents and transcripts are supplied to it and finished drafts are exported back. No back office, planning or record system integration is published. No meeting capture is offered, so transcripts come from elsewhere.

A competitor's published comparison advises firms to verify meeting capture and back office connectivity directly with the vendor, which points the same way. The founders present this as a deliberate choice, describing the product as built to work the way advice firms already operate rather than requiring them to adopt a technology company's workflow. That is a coherent position, and its cost is integration depth.

Delivery is a hosted service reached by uploading documents and downloading drafts. No hosting region or country is named, and no in country residency option is described. No data transfer mechanism is set out, and no subprocessor list exists. The gap matters because the upload architecture moves client files to the vendor's environment as a matter of course rather than as an exception. Where that environment sits is a first order question for any firm conducting due diligence. A competitor of comparable size publishes the processing country for each of its providers.

No rates, tiers or seat pricing are published, and volume bands, minimum terms and trial arrangements are not public either. For a tool sold to advice practices ranging from single adviser firms to larger groups, even the pricing unit is unclear, with no indication whether the charge is per user, per report or per firm. Every route resolves to a demo or contact request. Two direct competitors publish either a per user monthly rate or a public pricing page, so the absence reflects a choice rather than a norm in this segment.

Coverage is narrow in both senses that matter. The product serves one jurisdiction and one buyer type, United Kingdom financial advice firms, and within that it addresses one function, the drafting of suitability reports and annual reviews for advisers and paraplanners. The tool launched in February 2026 and one customer firm is named, so market presence is early rather than established. There is no deployment outside the United Kingdom, no adjacent institution type, and the document formats and regulatory expectations are built around a single national rulebook.