WealthAi

The removal test leaves a data aggregation layer and a marketplace, which is not the product. The assistant is described as agentic and acting rather than assisting, moving a user from request to result by handling retrieval, administration, reporting and workflows, with role-specific agents across compliance, suitability, client management, research, investments and operations.

The architecture is stated specifically as trained small language models paired with deterministic workflows, and a notetaker attends meetings, transcribes in real time and guides advisers through talking points.

Most agents are constructed to support rather than replace, with compliance agents monitoring, flagging and auditing for a compliance team, suitability agents supporting regulatory control, and the notetaker guiding an adviser through a conversation the adviser is having. Continuous monitoring replaces sample based review, which increases coverage rather than removing judgement.

Against that, the assistant is described as acting rather than assisting and the investments module extends through research and portfolio construction to trade execution, and no approval threshold, escalation route or human confirmation step is described for that path.

The architecture is itself a risk control and is disclosed with unusual specificity: trained small language models paired with deterministic workflows, which places rule bound processing where correctness is required and model judgement where interpretation is needed, and allows the system to update continuously without manual process rework. That is a more informative statement than most accuracy claims.

Held at B because the only published figure is a productivity gain of up to 70 percent from early adopters, with no accuracy, error rate, validation result or measure of how often agent output requires correction.

One named customer has taken a full front to back deployment across all users and agents, with specialist agents assigned to compliance monitoring, client management, research and operational workflows, which is materially stronger evidence than a pilot. A second firm's head of risk and compliance is quoted by name. A data partnership brings institution grade coverage from more than 650 institutions. Early adopters are reported to see up to 70 percent productivity gains. Held at B because the company is at pre-seed stage on around one million dollars raised, and the customer count remains small.

This is the clearest boundary statement in the index and it answers the exact question left open almost everywhere else: client data is never used to train models, and the system is stated to work exclusively for the firm deploying it. The commitment is backed architecturally rather than only contractually, since the platform is closed rather than routing through general purpose services, and the company frames unsanctioned external tool use as the risk its design exists to eliminate.

For a product ingesting meeting transcriptions, portfolio positions and client circumstances from firms that compete for the same families, that separation is the first question a prospective client would ask and it is answered before being asked.

The platform is described as closed, with the company positioning that architecture explicitly as protection against staff using unsanctioned external tools, which is a real risk in wealth management where client information is the asset. The commitment that client data never trains models is stated unambiguously. Standards are claimed to exceed the toughest regulatory and data requirements.

Held at B because no data processing agreement, retention schedule, subprocessor register or deletion commitment was located, and the platform holds meeting transcriptions and complete client records across custodians.

No attestation, certification, trust centre or enumerated framework was located, with security addressed through the closed platform architecture and a claim of exceeding the toughest regulatory and data standards. A private bank and a family office have deployed the platform across their whole operation, so security assessment has been passed, and the published control set that would let other regulated firms begin their own review does not exist.

Suitability appears as a named module rather than a generic feature, which matters because suitability is a specific and heavily examined obligation in advice, and the compliance product targets communications monitoring and audit directly. The company positions itself as built for regulated environments rather than retrofitted to them, and its data partner's status as an authorised account information service provider with European passporting is disclosed. Held at B because no regulator, rule or supervisory framework is named for the platform itself across the jurisdictions it serves.

The client population is affluent, which removes the access questions that dominate elsewhere in this index, and leaves a different exposure that is not addressed. Suitability determinations and portfolio recommendations are generated by agents, and a model trained on how advisers have historically categorised risk tolerance and objectives will reproduce those patterns, including any tendency to assess capacity for loss differently by age, gender or circumstance. Compliance agents flagging communications introduce a second question about what conduct is treated as anomalous. No testing, subgroup analysis or model behaviour disclosure appears.

No guarantee, indemnity or correction process was located. The firm is well served, retaining its own compliance oversight with the platform improving its coverage. The underlying client is not addressed: their meetings are transcribed, their circumstances recorded in an automatically maintained file, and suitability assessments generated from that record, with nothing stating whether they are informed that an agent attended and transcribed the conversation, whether they can see or correct what was captured, or how an inaccurate suitability determination is challenged.

The model class is named specifically as trained small language models rather than described vaguely, which tells a buyer something meaningful about how the system runs and is consistent with the closed platform commitment, since smaller models can be operated without routing data to a general purpose service. Seven data and analytics partners are named individually and the aggregation partner's regulatory status is disclosed. Held at B because no base model, provider or hosting arrangement is identified and no subprocessor list appears.

This is the deepest named integration set in the index. Seven external providers are identified individually across market data, research, economics, portfolio analytics and planning, delivered through a curated marketplace that removes the usual integration burden, and the data layer reaches more than 250 custodians and banks directly with over 650 institutions available through a named aggregation partner authorised as a payment services provider.

The company also states the platform connects with legacy systems rather than replacing them, requiring no migration project or consultant dependency, which is the constraint that stops most wealth firms modernising at all.

The platform is described as closed, which addresses isolation from external services without addressing location. No hosting provider, region selection, residency commitment or private deployment option was located, and for a product holding client records and meeting transcriptions for private banks and family offices whose principals may be resident in several jurisdictions, residency is a question those clients raise early.

More commercial structure is disclosed than most vendors offer, with no upfront investment, a stated low cost per user model, and an explicit argument that adoption pays for itself by allowing firms to retire existing vendor systems. The design is unusual in allowing individual users within one firm to select the tools suited to their role rather than being bound by a firm wide vendor decision, which implies genuinely per seat commercial terms. Held at B because no actual price, tier or marketplace revenue arrangement is published.

Buyers span wealth managers, family offices, private banks, asset managers and external asset managers, with the company stating the platform suits everything from solo advisers to global banks, and functional coverage runs across front, middle and back office including compliance, suitability, client management, research, portfolio construction, execution and operations. Coverage of the underlying data problem is genuinely global through custodian connectivity. The limit is deployment: the named firms are United Kingdom based and the footprint is European.