FinMate AI

The models carry the whole product and nothing sits underneath them. Transcription across virtual, telephone and in person settings, structured note generation against adviser specific meeting templates, drafting of follow up communication, population of forms from what was said, and inference of client mood across a conversation are all inference. Remove them and what remains is a recorder and a set of connectors with nothing to send.

The form filling capability is worth noting as the point where this goes slightly beyond documentation, because translating a spoken exchange into completed paperwork is an interpretation step rather than a transcription one.

Nothing acts without a person and no oversight design is published for the parts that could go wrong. Notes and tasks reach a client system through an explicit push rather than automatically, which is a real limit even though it is a workflow property rather than a control.

Searched for a confidence measure on a transcript or an inferred mood, any flagging of a passage the system was unsure of, an approval step before a generated form or message is used, or any audit record, and located none. One architectural fact cuts in this vendor's favour and belongs here: because it retains a recording, a firm that later doubts a note has a source to check it against, which is a fallback the non recording products in this pocket deliberately gave up.

Searched for a transcription accuracy figure, any measure of note completeness, a benchmark, a validation method or a revalidation cadence, and located none. Two capabilities in this product carry accuracy risk beyond ordinary summarisation and neither is measured.

Populating a form from a spoken conversation puts a machine derived value into a document a client may sign, and inferring a client's mood is a judgement about a person recorded in their file, presented as insight with no indication of how often it is right. Multiplying both across five languages, with no per language performance stated, compounds the gap.

Independent inclusion is plentiful and independent measurement is absent, and only the second carries this axis. The product appears as one of twelve advisor specific solutions reviewed in depth in a research firm's 2026 category buyer's guide, in an adviser network's own nine product comparison where reviewers recorded what they liked about it, and in competitors' category round ups.

That establishes it is a real product in the genuine consideration set, which is the corroboration test passed rather than an outcome. Searched for a named customer firm, an adoption figure, an independent ranking or share position, or any measured result, and located none. The one quantified claim, at least five hours saved per adviser per week, is self reported and attached to nobody.

An explicit and unambiguous commitment sits at the centre of this row, made in the first person and framed as a professional undertaking rather than a policy line, that the company will never train its models on customer data. It is independently corroborated: an adviser network's own product comparison separately recorded that no customer data is used for model advancement, which means a third party evaluating the product for its members satisfied itself of the same point.

That is stronger than an unverified assertion and it answers the central question of this axis directly. Held below the top grade because it is the only commitment made: searched for a retention schedule for recordings and transcripts, a deletion route, or any statement of what its transcription and language model suppliers retain, and located none.

Recorded as an unverified absence, and the exposure is larger than for its non recording competitors by design. This product captures and retains audio and video of client conversations, and accepts uploads of recordings made elsewhere, so a firm accumulates a library of recorded client meetings alongside the notes. That is a defensible choice with real benefits for verification, and it makes retention, deletion and access the questions that matter most.

Searched for a retention schedule, a deletion route, an access control model or published processing terms and located none in the material opened. Queued check: a privacy policy and processing terms.

A specific credential is named in the correct language and its provenance is one step removed, which places it between the two failure modes this index usually sees. The claim is a Type 2 service organisation control certification, and the wording matters because Type 2 is what a certified organisation says while adheres to leading standards is what an uncertified one says, so this passes the test that a comparable vendor in this pocket failed.

What it does not pass is the domain test: the statement appears in first person vendor authored copy hosted on a third party adviser directory rather than on the company's own material, and no trust page or security page was located. An independent adviser network review separately recorded that the product did not offer two factor authentication on its own system, relying on connected systems for that, though that review dates from 2024 and may be stale. Queued check: a trust page on the vendor's own domain.

A software supplier holds no licence and requires none, and no penalty attaches. The founder having practised as an adviser is a credibility signal about product design rather than any supervised status of the company. One consideration follows from what this product creates rather than what it is: recordings of client conversations and generated notes are examinable records for the advice firm, and searched for any statement about retention obligations, archiving destinations or how a firm evidences the provenance of a generated note to an examiner, and located none.

Searched for any fairness evaluation, any per language or per accent performance data, or any account of how client mood is determined, and located none. The mood inference is where this axis bites hardest. Reading emotional state from speech is culturally and linguistically variable, expressiveness norms differ substantially across the five language markets this product serves, and a client characterised as hesitant or dissatisfied has had a judgement about their disposition recorded in an advice file that shapes how the firm treats them. Nothing published describes what signals drive that inference or whether it behaves consistently across the populations the product was sold into.

Searched for a warranty, an accuracy commitment, a correction obligation or any allocation of responsibility between vendor and advice firm, and located none. The pre filled form is the sharpest exposure in this pocket for its size, because a figure transcribed wrongly from a conversation into a completed document travels further than a wrong sentence in a note: it can reach a provider, a custodian or a client signature.

The client has no described route to see the recording made of them, the note written from it, or the mood attributed to them, and the mood attribution is the one a person would most object to if they knew it existed.

Searched the vendor material, the directory listings and the independent reviews for a model provider, a model family, a version, a speech recognition supplier or an inference host, and located none. The five language capability makes the omission more consequential than usual, because multilingual transcription and generation at that range is very unlikely to be built in house at this company's size, so external suppliers almost certainly sit in the path and none is named. A commitment not to train on customer data covers what this vendor does with it and says nothing about which third parties the audio passes through on its way to a transcript.

Around eleven systems named individually across three categories, covering the client relationship systems independent advice firms actually run, the major conferencing platforms and both dominant calendar services. An independent adviser network review singled out the push of tasks from generated summaries into one of those client systems as the standout, which is corroboration of write depth rather than mere connection.

File upload of externally captured audio is a useful fourth route for firms whose recordings originate elsewhere. Held below the top grade because no custodian, planning platform or programmatic interface was located, and because the enumerated list comes from a third party review dated 2024, so it establishes what existed then rather than what exists now. Queued check: the current integrations page.

Recorded as an unverified absence, and residency is a genuine question rather than a formality for this vendor. Operating in five languages implies European users, and European advice firms handling client recordings face residency expectations that a United States hosted service may not meet. Searched for a hosting region, a residency commitment, any European data location, or any account of deployment topology, and located none.

A per user monthly band circulates in third party review material with enterprise pricing noted as available, and this index does not grade commercial terms from an aggregator. Searched the vendor's own material for a published rate, a tier structure or a trial arrangement and located none. Queued check: the vendor's pricing page.

One buyer type and the widest geographic reach in this pocket, which is an unusual combination worth recording precisely. Every comparable meeting assistant assessed in this session operates in a single country, either the United States or the United Kingdom.

This one ships in five languages spanning English, Spanish, German, French and Italian, which addresses continental European and Latin American advice markets rather than gesturing at international availability, and the company has engaged the Australian advice profession through its trade media. Held below the top grade because the buyer remains a financial advice firm and nothing else, with no bank, insurer or asset manager segment addressed, and because no firm size range is stated.