WealthAi

Specialist agents run work across the front, middle and back office, from meeting preparation to custodian management. Each workflow is routed between AI models and deterministic logic according to its risk.

Human review is required on the workflows, and the system records who reviewed each output, when they acted and what they decided. What isn't published is the threshold, if any, at which an agent may act without that review.

Every answer cites its sources, and every action lands in a tamper evident log, so a firm can reconstruct what an agent did. What WealthAi doesn't publish is any accuracy, error or validation figure, so there's no way to tell how often the agents get it wrong.

Six clients are named, including Mediobanca, Patronus Partners and Saranac Partners, and staff at five of them are quoted. The quotes speak of efficiency and hours saved, but none of them carries a figure.

Client data stays out of shared model training, and each firm's data is isolated at the row level. Prompts, documents and client information run in controlled model environments under set handling rules, which answers the question competing firms ask first.

Article 28 processing agreements are in place, and each firm sets its own retention window, up to 365 days, while remaining the data controller.

ISO 27001, SOC 2 Type II and Cyber Essentials Plus are all still in progress, leaving a vendor review with no report to read today. An annual penetration test by a CREST accredited firm is already in place.

The client firm stays the regulated entity and decision maker, and WealthAi says so plainly while mapping the platform to UK GDPR, the Swiss data protection act and FINMA related requirements. Its agents also handle market abuse and transaction reporting, which puts the product inside duties a regulator examines.

Suitability reviews, vulnerability assessments and dissatisfaction detection are exactly the outputs where uneven treatment of clients would show up. Testing across groups such as age or gender isn't described.

Meeting Intelligence records client meetings, yet a client has no published way to see or correct what it captured, and WealthAi publishes no commitment for when an agent makes a mistake.

Who processes client prompts is the open question. The language models behind the agents go unnamed, and the subprocessor register is kept private.

Eight providers feed a marketplace of data and tools, among them Morningstar, Flanks and Bridgewise. The connections to CRMs, portfolio systems and custodians are claimed without a single system named, and those are the ones a wealth firm runs on.

Data lives in Azure's UK South region, with recovery and backups in AWS Frankfurt, and UK and EU residency are both supported. A Swiss region is planned but not yet live.

Basic is priced at 250 pounds per user a month, with a free tier and 14 day trial below it and quoted Enterprise pricing from ten users up. That per seat figure lets a firm size the cost before a sales call.

Buyers range from advisory practices and family offices to private banks. Most named clients are UK firms, and outside Britain the list stops at one Italian bank.