Finster AI

The company describes itself as an AI-native intelligence and agent orchestration platform and the architecture supports the claim, combining a proprietary data pipeline grounded in verified corporate data with language models rather than relying exclusively on the models alone. Agents synthesise structured data, unstructured content and institutional knowledge into research outputs, briefing decks and monitoring. The founder is a former researcher at a leading AI laboratory and the team is drawn from two AI labs alongside two investment banks.

Output is draft material a professional finishes, covering memos, research, client materials and briefing decks, and the stated aim for asset managers is to let teams focus their time on investment judgement rather than on synthesis.

One framing is genuinely distinctive: rather than producing conclusions, the platform highlights what is changing, what is inconsistent and what is underexplained, and turns that into sharp questions for use in diligence, earnings calls or management meetings, which leaves the judgement with the analyst by design. Held at B because no review requirement or approval step is published for material that reaches clients.

The response to hallucination is structural rather than promissory, grounding a proprietary pipeline in verified corporate data instead of relying exclusively on standard language models, and outputs carry source citations so a claim can be traced to a filing or transcript rather than accepted.

Traceability is stated repeatedly as a design property and the company frames the requirement plainly, that front-office users have zero tolerance for fabricated content in client communication, investment documents and market research. Held at B because no accuracy measure, error rate or evaluation methodology accompanies the architecture.

Customers are characterised precisely without being named: tier-one global investment banks and buy-side asset managers collectively managing over 800 billion dollars and employing more than a thousand financial analysts, which is a usable measure of deployed scale. A major Swiss bank has taken a strategic investment position, and a leading market data provider is a named development partner integrating the technology into its own banking platform. Held at B because no institution is identified as a customer and no outcome, productivity or accuracy figure is published.

No boundary statement about the vendor's own use of customer material was located. Private intelligence is claimed as a feature and material nonpublic information controls are described, both of which concern isolation between the customer's users rather than between the customer and the vendor. Nothing states whether institutional knowledge ingested from one bank informs models serving another, which matters acutely where the customers are competing deal teams.

The confidentiality problem is named specifically rather than generically: handling material nonpublic information requires strict controls and robust governance, and the platform is built to support workflows sensitive to it, with private intelligence stated as a product property alongside personalisation. That is the correct frame for this buyer, where information barriers are a legal requirement rather than a preference. Held at B because no data processing agreement, retention schedule or subprocessor register was located.

No attestation, certification, trust centre or enumerated control set was located. Tier-one investment banks have completed supplier assessment before allowing the platform near confidential deal material, and a major bank has invested, so assurance exists privately while nothing is published for a prospective buyer beginning its own review.

The governing constraint is named at the right level of specificity for this buyer, with material nonpublic information handling treated as the design requirement and deployments described as maintaining strict regulatory and information controls within front-office operations. That is the concept insider dealing and information barrier rules actually turn on. Held at B because no regulator, statute or research and communications rule is named, and outputs reaching clients sit within marketing and record-keeping obligations that go unmapped.

No fairness testing or governance disclosure was located. The adapted exposure runs through analytical framing rather than individuals: a platform that decides what is inconsistent or underexplained across a coverage universe shapes which companies receive scrutiny and which questions get asked in earnings calls and management meetings, and nothing describes how that selection is constructed or reviewed.

No guarantee, indemnity or correction process was located. The exposure travels outward, since a briefing deck or investment memo containing a model-introduced error reaches a client as the bank's own work product, and nothing describes what the vendor stands behind or how an error traced to synthesis rather than authorship would be handled.

The data dependency is disclosed unusually well, naming four premium providers covering market data, fund research, private markets and company intelligence alongside public filings and transcripts, so a buyer can see what the research is actually built from. The architecture is stated as not relying exclusively on standard language models, which discloses the shape of the model dependency. Held at B because no base model, provider or version is identified behind the generative layer, so the model half of the chain remains unnamed.

Four premium data platforms are named as ingested sources alongside regulatory filings and earnings transcripts, and one of them is additionally a development partner whose banking AI platform provides the environment for workflow automation, which is a deeper relationship than a data licence. The company states outputs land within existing workflows and tools rather than a separate destination. Held at B because no order management, research management or client relationship system is named individually.

No hosting provider, region, residency commitment or private deployment option was located. That gap is notable for a platform handling material nonpublic information across a London headquarters and United States operations, where the location of processing bears directly on the information barrier controls the product is built around.

No pricing, packaging or basis of charge was located. The company describes deployment as requiring true partnership and deep workflow understanding rather than an easy button, which signals an implementation-heavy engagement without indicating what either the platform or the implementation costs.

Three segments are addressed with separately described workflows, covering investment banking where speed to a deal-ready output matters, asset management where coverage breadth is the constraint, and private credit. Operations span a London headquarters and a New York build-out serving United States clients. The company is explicit that it tailors to individual roles, workflows and information needs rather than offering one interface. Held at B because presence is evidenced only in aggregate.