CogniCor

The removal test leaves a set of connectors and some templates. Every function is model work: ingesting a firm's enterprise documents and answering from them, preparing for meetings, transcribing and summarising client conversations, drafting follow up correspondence, populating relationship management fields, and surfacing gaps and opportunities from combined planning, portfolio and interaction data. The company's origin reinforces the position, since the platform grew out of its chief executive's doctoral research rather than from a workflow tool that later added a model.

The boundary is stated in the company's own words and it is drawn at the right place: automation runs up to last mile approvals and exceptions, with relationship record updates applied where enabled, so the firm controls which write actions the machine may perform and a person handles the final step. That is a clearer division than most productivity copilots offer. What is not described is the perimeter around consequential actions.

The company's own illustrations include instructing the assistant to set up a required minimum distribution for a client or initiate a beneficiary change, both of which move money or alter who inherits it, and nothing published states what approval those specifically require or how a mistaken instruction is caught before execution.

One real control is described and no measurement accompanies it. Copilots answer from the firm's own ingested enterprise documents rather than from general knowledge, which is retrieval grounding and the standard architectural defence against fabrication, and it means an answer should be traceable to a source the firm supplied.

Against that, accuracy is asserted rather than shown, with the platform described as giving compliant and accurate answers and no error rate, evaluation result or citation mechanism published. For a product whose output includes client summaries and drafted correspondence going to households, an unmeasured fabrication rate is the material gap.

Independent recognition is unusually dense and comes from bodies that assess this specific market: two consecutive industry awards in the artificial intelligence category from the leading wealth management trade publication, an impact award from a financial services research firm, a finalist place in an adviser technology awards class, first place in an international competition judged across three thousand entrants, and an invitation from a national monetary authority to speak at its financial technology festival.

Customers are described rather than named, as leading wealth management firms, large broker dealers and registered investment advisers, with one segment characterised by assets above eight billion dollars, and testimonials are quoted anonymously. Distribution through two major enterprise marketplaces indicates the platform has passed each vendor's listing requirements. No deployment count, user count or quantified outcome was located.

The architecture splits in a way worth stating precisely, because one half answers the boundary question and the other opens it. Per customer, copilots are grounded in that firm's own ingested document library and learn its products and services from its own material, which contains the knowledge base to the institution that owns it.

Underneath that, the platform itself is described as trained on more than 200,000 enterprise documents, and nothing states whose documents those were, whether they came from customers, or whether a firm's ingested library contributes to that base for others. A wealth firm's product documentation and procedures are competitively meaningful, so the question is not academic.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The payload is the full picture of a client household, drawing planning documents, portfolio holdings, relationship management records and the content of client meetings into a single continuously updated context layer, which is precisely the aggregation that makes the product useful and the concentration that makes it sensitive. Nothing published states how long meeting content persists, what happens to a household record when the client leaves the firm, or what the client is told about the processing.

No attestation, certification, trust centre or enumerated framework was located. Listing on two major enterprise marketplaces implies passing each operator's technical and security review, which is a real if indirect signal, but it is not an artifact a buyer can read and it covers the listing rather than the customer deployment. For a platform aggregating household financial data and meeting content for regulated advisers, a published assurance set is what the firm's own compliance function will require.

No supervisor, statute or instrument is named. Compliance appears as a product feature, with assistants offering guidance on compliance processes and forms and the platform described as producing compliant and accurate answers, which is a claim about output quality rather than an identification of the rules being satisfied.

The gap is specific given the illustrated use cases, since required minimum distributions and beneficiary designations are governed by detailed retirement account rules with their own deadlines, penalties and documentation requirements, and a copilot operating in that territory without naming them leaves a buyer to establish the fit alone.

No consumer credit or identity decision applies, so the axis adapts as it did for Zeplyn and Avantos. The governance question is attention allocation. The insight engine surfaces household level signals, gaps and opportunities and triggers next best actions, which determines which clients an adviser contacts and what they are offered, and any ranking built on portfolio size or revenue potential will systematically direct adviser time toward wealthier households, which is rational commercially and has distributional consequences for everyone else on the book. Nothing published describes how opportunities are prioritised, whether the model is optimising for client outcome or firm revenue, or whether a firm can inspect the ranking logic.

No guarantee, indemnity or falsifiable commitment was located. Accountability is allocated sensibly by the stated automation boundary, since a person handles last mile approvals and exceptions, so a regulated adviser remains answerable for what reaches the client and the ordinary complaint and arbitration routes apply regardless of how a document was drafted. That is the regime working rather than the product providing recourse.

Nothing describes correction of an inaccurate client summary already sent, notification when an ingested document was misread, or any route for a client affected by a copilot generated error they were never told about.

Three links are named where most vendors name none. The frontier model service powering the digital assistants is identified explicitly by provider and product, the hosting environment is named as the same provider's cloud, and distribution runs through that cloud's marketplace and a major customer relationship platform's application exchange. The training base is quantified at more than 200,000 enterprise documents, which describes scale even though it does not identify origin. What separates this from an A is the commercial layer: no retention or training terms with the model provider are stated, which is the step Marloo takes, and no subprocessor list appears.

The positioning is to disappear into the firm's existing stack rather than to add another application, stated directly as your platform, no third party apps, no context switching, and the adviser product is published on a major customer relationship platform's application marketplace with the ability to update records in place. Data is drawn across planning, portfolio, relationship and interaction systems into one household view, which is the integration that makes the product work.

What is not published is the specific list. In wealth management a handful of portfolio accounting, financial planning and custodial platforms hold the field, and none is named, so a firm cannot confirm its own stack is supported without asking.

The hosting platform is named, with the digital assistants stated to be hosted in a major public cloud and offered through that cloud's marketplace, which is more disclosure than most vendors here provide. It stops short of a residency position: no region selection, data location commitment or private deployment option is published, and nothing indicates whether a firm can require its household data to remain in a particular jurisdiction. Same position as Symend, where the cloud is named and the residency is not.

No pricing, packaging or basis of charge was located on the company's own material. The product is listed on two enterprise marketplaces, which often carry transactable pricing, so a rate may exist in those catalogues even though the vendor does not publish one itself. Nothing indicates whether charge falls per adviser, per household, per copilot or as a platform fee, which matters for a buyer set ranging from a small registered adviser to a national broker dealer.

Two adjacent industries are served on one platform, wealth management and insurance, with buyers spanning registered investment advisers, large broker dealers, insurers and their agents, and the company's earlier work reached banks and support organisations.

Within wealth the coverage follows the whole relationship lifecycle rather than one moment, from prospecting and planning through onboarding, client review meetings and ongoing management, and the household context layer is built to span planning, portfolio and relationship systems together. The limits are that this is adviser productivity rather than the institution's wider operations, and no footprint outside one market is evidenced.