Automwrite

Generative document production is the entire product. The platform transcribes meetings, extracts structured client facts from unstructured conversation and uploaded files, identifies missing information and conflicts, and writes a full suitability report in the firm's own template and voice. Remove the models and what remains is a template library, a file uploader and an electronic signature step, which is not a sellable product in this market and not what any customer is paying for.

The company describes the shift explicitly as moving from rule based workflows to model driven generation, and its own founder essay discusses building machine learning algorithms before the product existed.

A gate is named and positioned in the execution path. Compliance gated checks run to confirm that all necessary information is present before a report is written, which places a control ahead of generation rather than after it. The workflow then routes through explicit review, with the adviser finalizing before anything reaches the client.

The company writes publicly about managerial accountability for automated outputs under the senior managers regime, arguing the model removes extraction work rather than responsibility. The gates are asserted rather than described. Nothing states what the checks test for, how many there are or what happens when one fails. Nor does anything say whether a firm can configure them, or whether any sampling of approved output occurs.

Accuracy claims are made and none is measured. The company states its proprietary technology enables unparalleled accuracy and that the platform produces consistent, replicable outputs of human grade quality, but publishes no error rate, no evaluation methodology, no test set description and no independent assessment.

The compliance gated checks run before generation and test whether required inputs are present, which is input completeness rather than output correctness, and the two are easy to conflate. Nothing describes how the system behaves when source material is thin or contradictory, how often a generated report requires substantive correction rather than editing, or how model changes at the named third party providers are detected and managed. For a product whose output is the primary regulatory evidence on an advice file, an unmeasured accuracy claim is the central gap.

Five customer firms are named on the vendor's own site, each with a named individual and their title. They span a national retirement advice business, an international wealth group, a small practice and two advice firms. The one quantified figure is prospective rather than realized. The chief executive quoted says report production previously took up to two hours and that with the automation the firm expects to bring it under a minute.

An expectation stated before deployment is a forecast, not an outcome. The remaining quotations describe ease of adoption and responsive support rather than measured results, and no independent party with anything at stake has assessed the product.

The stewardship half is answered directly. The transparency page states that third party model providers are not permitted to use customer data to train or improve their models. That is an affirmative boundary statement, and it matters because the material flowing to those providers is recorded client meetings and the full financial circumstances of retail investors. Data residency per provider is also published. The safety half is thinner.

Compliance gated checks confirm input completeness before generation, but nothing describes evaluation of output quality or controls against fabricated content in a document that becomes regulatory evidence. Nothing says what happens when the source material is ambiguous. The training commitment is also qualified by a clause deferring to each provider's own policies.

Automwrite publishes a dated data protection policy as a standalone document. It sits alongside a privacy policy, terms, cookie controls and the AI transparency page, which together address the specific question this product raises: what happens to a recorded client conversation and the financial circumstances extracted from it. The stated position that third party providers may not train on customer data is the most material privacy commitment in the set. Gaps remain.

No privacy management certification is held, and no retention or deletion period is published for meeting recordings, transcripts or generated documents. No subprocessor list exists beyond the AI providers, and the training commitment defers to each provider's own policies rather than standing alone.

There is no published information security management certification or SOC report, and no trust center or security scope statement. A dated data protection policy is published as a downloadable document, but a data protection policy is a privacy artifact rather than a security attestation, and the two should not be read across.

This matters in proportion to what the platform holds: recorded client conversations, transcripts and complete financial circumstances for the retail clients of regulated advice firms. Buyers in this segment increasingly face due diligence questionnaires from their own networks and compliance providers, and nothing published would answer one.

The product is built around a specific regulatory artifact rather than positioned near one. Automwrite holds no financial license, which is correct for a technology supplier. The suitability report is the primary file level evidence of compliance under the conduct rules. The company writes directly against the relevant sourcebook provision, the consumer duty regime and the senior managers regime, including published commentary on where accountability sits when an automated output is wrong.

That is substantive regulatory grounding rather than a compliance badge. Still, no regulator has assessed, tested or supervised the product, and no authorization or program enrollment exists. The description of outputs as compliant is the vendor's own characterization.

Nothing published addresses whether the system treats clients evenly. The risk in this product is specific and not hypothetical. A model drafting suitability rationales from recorded conversation may render the same circumstances differently depending on how a client speaks, their accent as handled by the transcription layer, their fluency or their age. The resulting document is the file evidence that the advice was appropriate for that person.

Transcription accuracy across accents is a documented source of disparity, and this platform depends on a third party speech provider for it. No demographic analysis or bias testing is published, and there is no evaluation of transcription performance across speaker populations or governance framework covering these questions.

Responsibility is asserted to sit with the adviser and nothing sets out what happens when the document is wrong. The company argues correctly that professional judgment and accountability remain with the regulated individual under the senior managers regime, which is the right legal position, but that framing resolves the vendor's exposure rather than the client's. No accuracy guarantee, remediation commitment or liability allocation is published.

The affected party is a retail client who receives a report explaining why a recommendation suits them, is not told a model drafted it, and has no route to establish how the reasoning was produced if the advice is later disputed. For a document that exists specifically to evidence that advice was suitable, the absence of any recourse position is a substantive gap.

A dated public transparency page tabulates every third party model provider the platform relies on. It names each company, describes the specific processing activity it performs, and states the country in which that processing occurs. Two named LLM providers cover reasoning and content generation, a named speech provider covers transcription of meeting audio, and a named infrastructure provider covers meeting capture. The company commits to updating the page as providers change.

An institution can therefore see which providers sit behind an output. Two qualifications remain. No model version identifier is published, so a firm still cannot tie a specific report to a specific model build. The commitment that providers may not train on customer data is hedged with a clause deferring to each provider's own policies rather than stated flatly as a contractual bar.

Integration is real at the productivity layer and absent at the systems layer. The meeting bot supports the three major conferencing platforms by name, and native applications ship on both mobile stores. Electronic signature is built into the delivery step, and the platform includes its own client record system. What is missing is the integration that matters for this buyer.

No adviser back office or planning system is named in the published material, though those platforms are where a United Kingdom advice firm's client records actually live. A customer testimonial refers to integration with an existing system without naming it. At least one direct competitor names four such platforms explicitly, so the omission is a comparative gap rather than an unreasonable expectation.

Residency is enumerated where it actually matters for this product. The transparency page states the processing country for each third party model and transcription provider individually. Language model and speech processing sit in the United Kingdom, and meeting capture infrastructure in France.

For a United Kingdom advice firm handling client data under domestic data protection rules, knowing that generation and transcription of client conversations occur in country is a substantive answer rather than an assurance. Delivery is hosted software reached through a browser, alongside native applications on both mobile platforms. Limits remain.

The disclosure covers the AI subprocessors rather than the platform's own primary hosting, and no region is stated for the customer data held in the product itself. No residency option or configuration is offered to the customer.

The commercial model is self serve and openly presented, in a category where almost every route ends in a sales conversation. A dedicated public pricing page sits in the primary navigation, and a three day free trial is offered with sign up directly from the site. The company states a firm can onboard itself in two minutes without assistance. That combination tells a buyer the product is bought rather than sold and that evaluation carries no commitment. The specific rates, tiers and any volume or seat structure are not reflected here, and nothing indicates whether larger firms move to a negotiated arrangement.

Coverage is narrow rather than merely undocumented. The product serves one jurisdiction and one buyer type: United Kingdom financial advice firms. Within that segment the range is real. Named users span a national retirement advice business, an international wealth group and single adviser practices, and the company states it is used by solo advisers, regional firms and larger advice groups.

There is no deployment outside the United Kingdom and no adjacent institution type such as banks, insurers or discretionary managers. The regulatory framing is built entirely around one national rulebook, which would need rebuilding for any other market.