Jump

The models carry the entire throughput and the assistant native reading applies cleanly. Capture, transcription, structured note generation in the adviser's own voice, task extraction, financial data extraction, recap email drafting and the retrieval layer that answers questions across a book of past conversations are all inference. Strip the models out and what survives is a calendar connector and a set of write paths into other people's systems, with nothing to write.

That is the opposite of the legacy platform reading, where a workflow spine predates the models and outlives their removal. The scheduling conveniences the product also offers, such as conflict handling and reminder sending, are peripheral rather than the thing being bought.

The oversight design is published, specific and configurable by an administrator rather than described as a principle. Every generated note, task and record update is presented for review and edit before it synchronises anywhere, which places a human decision between generation and the client record.

Above that sits a configurable attestation: a firm that requires sign off before data leaves the platform can set a prompt requiring the adviser to confirm they have reviewed each output, and enterprise audit trails record what happened. Capture itself is a further control, since the firm selects transcript only, summary only or full recording to match its supervision policy. That combination of a gate before every write, administrator configuration of the gate, and an audit record behind it is what separates this grade from vendors that merely assert human involvement.

The pipeline is explained and the performance of that pipeline is unmeasured, and the distinction matters because only the second is what this axis asks about. Published support material describes the architecture plainly, being speech to text transcription followed by generative summarisation into notes, tasks and drafts.

Searched for a transcription accuracy figure, any measure of summary fidelity, a benchmark, a validation method, an account of behaviour on unfamiliar terminology, a drift statement or a revalidation cadence, and located none. The unmeasured risk is specific to what this product is for: a summary that omits a material client instruction is wrong in a way the adviser cannot detect, because the adviser reads a clean and complete looking note and never learns what was left out, and that note is then synchronised into the record a compliance reviewer will later rely on.

Corrected upward within the same session, and the reason for the correction is worth recording because it turned on a source found after the first grade was cut. This row was initially held one grade lower on the reasoning that the single quantified customer figure, a reported 83 percent reduction in post meeting tasks at a named firm, sits beside a footnote calling the scenario hypothetical, and that the strongest independent publication located was a distribution partner rather than a customer.

Both observations still stand. What overturned the grade is independent research: a widely followed adviser technology researcher, publishing its own periodic survey of adviser productivity and technology adoption rather than any vendor's material, identifies this vendor as the clear leader of the meeting support category at an adoption rate approaching one in ten of all advisers surveyed, with the nearest competitor a distant second.

A separate independent software survey recorded the category reaching roughly 43 percent penetration of advisory firms in its first year of measurement. An adoption figure measured across an entire profession by a third party with no commercial interest in the result is stronger operational evidence than any single case study, because it cannot be selected. Named advisers quoted with professional credentials sit alongside it. The disclaimed 83 percent figure earns nothing and is recorded here so a reader knows it was seen and set aside.

The central question this axis asks is answered directly and without hedging, which is uncommon enough to be the reason for this grade. Published support material states plainly that customer data is not used to train models, and goes further than the assertion by explaining the reasoning: the underlying language models are pre trained, so customer content is not required for training or retraining, and the vendor cites supervisory caution about client personal information being retained inside models where it could later surface.

Alongside that sit customer controlled retention with manual and automatic deletion, and a capture choice that lets a firm avoid creating an audio artefact at all by selecting transcript only or summary only. A firm can therefore establish what is created, how long it lives, and that it does not become training material. The remaining gap, and it is worth naming, is that these are commitments about the vendor rather than about what its transcription and language model suppliers retain.

Concrete and checkable commitments are published across the areas that matter for a product holding recorded client conversations. Encryption is stated for data in transit and at rest, multi factor authentication is stated for access, and the firm rather than the vendor controls what is retained, with both manual and automatic deletion available and the stated purpose of accommodating the firm's own retention and books and records policy.

Meeting data is described as retained until the customer chooses to remove it, which is a clear allocation of control even though it is not a retention schedule. Held below the top grade because the material located is a support article rather than a data processing position: searched for processing terms available to a customer, a subprocessor list, a lawful basis account or a published privacy policy governing the client whose conversation is captured, and located none of it in primary material.

Four separate elements are present and together they clear the bar this index sets. A specific attestation is named rather than gestured at, being a Type II report under the service organisation control framework, obtained through regular audit rather than described as alignment. Continuous control monitoring runs through a named compliance automation platform. Third party penetration testing is stated as recurring.

A dedicated trust centre is published on its own subdomain, hosted by that same named platform, which is the presentation shape this index treats as the reference positive. Encryption in transit and at rest and multi factor authentication are stated alongside.

One honest limitation on this record: the trust centre is a client rendered application and its contents did not load on fetch, so the certification list and any further standard held are asserted from the vendor's support material rather than read at the register. Queued check: open the trust centre for additional standards and the subprocessor list.

A software supplier to advice firms holds no licence and requires none, and no penalty attaches to that. The position is worth describing precisely because the product is built harder against its customers' regulatory obligations than most in this index: retention controls are described as accommodating the firm's books and records policy, capture modes are matched to supervision requirements across adviser and broker dealer firm types, and the generated output is presented as suitable for compliance documentation.

Every one of those is a property of the customer's obligation, discharged by the customer. The vendor itself is unsupervised and unexamined, and nothing published describes what an examiner reviewing a firm's records would be shown about which entries were machine generated.

Two distinct exposures exist here and neither is addressed in published material. The first is transcription: accuracy varies with accent, dialect and speech pattern, a problem this pocket clearly knows about because a competing vendor markets regional dialect handling as a named feature, and a systematically poorer transcript produces a systematically poorer advice record for the clients affected.

The second sits in the insights capability, which reads sentiment, impactful topics and recurring patterns across a whole book of clients: a system that decides which relationships merit attention is allocating adviser time, and a consistent tendency in what it surfaces directs service toward some clients and away from others without anyone choosing that. Searched for evaluation across speaker groups, any fairness testing, or any account of how insights are ranked, and located none.

Searched for a warranty, an accuracy commitment, a service level, a correction obligation or any allocation of responsibility between vendor and advice firm, and located none. The sharper observation is that the product's best oversight feature is simultaneously its liability mechanism: the configurable attestation, in which an adviser confirms having reviewed each output before it synchronises, is a genuine control and is also a logged act transferring responsibility for the resulting record from the vendor to the person who pressed the button.

The vendor's practice of footnoting its own outcome claims as hypothetical points the same way. The party with the least recourse is absent from all of it, since the client whose conversation is recorded, summarised and filed has no described route to see the record made of them or to contest it.

The category of supplier is described and no supplier is named, which is the common shape and is notable here only because of the company this vendor keeps on its other disclosure rows. Support material refers to speech to text technology and to large language models generically, and explains that those models are pre trained elsewhere, which concedes an external dependency without identifying it.

Searched the product pages, the integrations directory and the security and privacy support material for a provider name, a model family, a version, an inference host or a subprocessor list, and located none. A vendor that publishes a residency statement, an attestation and an explicit training isolation commitment has clearly decided disclosure is worth doing, which makes the silence at the model layer a choice rather than an oversight. Queued check: the trust centre, where a subprocessor list would normally sit.

This is the strongest integration estate located in the adviser assistant pocket and it is the axis on which the vendor most clearly separates from its peers. More than forty two way integrations are claimed and a substantial number are named individually across distinct system categories: client relationship systems, financial planning tools, portfolio management and reporting platforms, a major custodian, telephony and document storage.

The connections run in both directions, pulling client context into meeting preparation and pushing notes, tasks and record updates back out. The depth claim is specific rather than general: outputs can be mapped to custom fields, custom objects, events and workflows rather than to notes and tasks alone, contacts are matched automatically and content is formatted to the destination schema, with the mapping configured once and applied to every subsequent meeting.

An actual residency statement is published, which is rarer in this index than it should be and is distinct from the far more common practice of naming a cloud host and leaving location unstated. Data is stated to sit on servers in the United States, specified to state level. That is a location a buyer can act on rather than an infrastructure relationship a buyer has to infer from.

Held below the top grade because a single stated region is the whole of it: searched for a residency choice, a second region for customers outside the United States, a dedicated or single tenant option, or any account of deployment topology, and located none. For a product marketed only into a United States buyer base that is a coherent position rather than an omission, but it is a narrower disclosure than the top grade describes.

A per seat figure for a named tier circulates in third party review material and this index does not grade commercial terms from an aggregator, a rule earned when three directories returned three different prices for one vendor and one of them was four years stale.

What is established from primary material is structural rather than numerical: a free trial exists, and advisers reached through an aggregator platform partnership receive enterprise pricing negotiated by that platform rather than the list arrangement, which means at least two commercial routes exist with different terms. Searched for a published rate, tier structure or seat minimum on the vendor's own material and located none. Queued check: the vendor's own pricing page.

Breadth here runs across firm type and firm size within one profession rather than across institution categories. The published buyer set spans sole practitioners and independent advisers, registered investment advisers, independent broker dealers and hybrid firms, and reaches enterprise scale through platform partnerships where an aggregator makes the product available to its affiliated advisers under enterprise pricing.

A stated base of more than 35,000 advisers indicates the range is real rather than aspirational. Held below the top grade for two reasons: the material located describes a single jurisdiction, with hosting, record keeping language and regulatory references all pointing to the United States, and the buyer is a wealth advice professional in every case, with no bank, insurer or asset manager segment addressed.