Farseer

The product is intelligence extraction and nothing survives its removal. A proprietary cloud engine combines search, text analytics with specific strength in Chinese language processing, knowledge graphs and machine learning, with generative capability layered on, applied to trillions of alternative data points from global news, social media and capital markets databases. The Chinese language specialisation is genuine domain engineering rather than a general model pointed at a regional market.

The system informs rather than decides, producing research dashboards, real-time alerts and risk scans that analysts act on, and two features leave meaningful control with the client: sentiment weightings are user defined rather than fixed by the vendor, and thematic indices are co-designed with the client rather than supplied. Held at B because no confidence measure, uncertainty range or escalation guidance accompanies alerts that feed investment and risk decisions.

Techniques are named at architecture level, covering text analytics, knowledge graphs and machine learning, and no accuracy measure, validation method, backtest or error rate is published for any of them. For alerting that feeds portfolio risk decisions, false positive and false negative rates are the numbers that matter and none appear, nor is there any description of how the engine is evaluated as language, sources and market conditions change.

More than 50 listed companies and securities houses across Hong Kong and the mainland are stated as clients, and one is named and significant: the territory's exchange operator, which is an unusual reference for a company of this size. A major Chinese technology group is among its investors and it was in the first cohort of an artificial intelligence lab that group co-founded, with a financial media group and a leading university finance school as partners. Held at B because no outcome, retention or performance figure accompanies the client count and no deployment is described in detail.

No boundary statement was located, and one aspect of the business deserves recording plainly rather than being smoothed over. Alongside investment analytics, the company markets a public governance line to government agencies that senses shifts in public sentiment, quantifies communication gaps on policy issues and tracks the spread paths and source attribution of information judged to be misinformation.

That is a materially different activity from investment research, it runs on the same engine, and nothing describes what separates the two, what data flows between them, or how client material is isolated.

No data protection agreement, retention schedule or subprocessor list was located. The platform ingests social media and news at scale and performs financial crime screening, both of which involve personal data about individuals who are not the customer, and a separate product line tracks public sentiment and attributes the spread of information, which raises handling questions the published material does not reach.

No attestation, certification, trust centre or enumerated control set was located. An exchange operator and more than fifty securities houses have presumably conducted their own supplier review before connecting, so assurance exists privately, and nothing is published for a prospective client beginning that process.

Investment research is framed explicitly around the European research unbundling regime, which is the specific rule that reshaped how research is paid for and therefore why automated research has commercial value, and the product also claims full coverage of environmental disclosure requirements and cross-border regulation. Financial crime screening is offered as a risk management function. Held at B because no local regulator or licensing position is stated for a company serving brokers and an exchange, and the cross-border claim is asserted without naming which regimes.

No fairness testing, methodology disclosure or governance statement was located. Sentiment weightings being user defined is a transparency feature of a kind, since the client sets the interpretation rather than inheriting the vendor's, and it also means the resulting scores carry no independent standard. For a platform that assigns sentiment to companies and management, and separately measures public opinion for government clients, the absence of any published methodology is the substantive gap.

No guarantee, indemnity or correction process was located. The affected party here is the company or executive scored, since risk scanning covers management as a named category and sentiment is assigned from news and social sources, and nothing describes whether a subject can see an assessment, contest an inaccurate signal, or correct source material that has been misattributed to them.

The engine is described as proprietary, which discloses that the capability is built rather than licensed, and no base model, provider or hosting arrangement is identified behind the generative layer. Data inputs are characterised only by category as global news, social media and capital markets databases, with no source, vendor or coverage detail, so a client cannot establish what the intelligence is actually derived from.

Delivery is explicitly dual, offering both a hosted platform and programmatic access so intelligence can flow into a client's own research or risk systems rather than remaining in a separate dashboard, with web and application access alongside. Held at B because no order management, research management, portfolio or market data system is named individually and no developer documentation was located.

Delivery is described as a proprietary cloud with no hosting provider, region option or residency commitment stated. That gap matters more than usual for a platform serving clients across Hong Kong and the mainland, where cross-border data transfer rules are strict and an exchange operator or securities house would need the question answered before onboarding.

No pricing, packaging or basis of charge was located for either the cloud product or interface access. Given delivery spans dashboards, alerting, applications and programmatic access with client-specific configuration, the charging basis could follow seats, queries, data volume or index co-design work, and none of it is indicated.

Buyer coverage is genuinely varied for a company this size, spanning securities brokers, institutional investors, asset managers, stock exchanges, listed companies and government bodies across Hong Kong, mainland China and the wider region. Use cases run across investment research, risk scanning, environmental and governance analysis, financial crime screening and investor relations. Held at B because the footprint is concentrated in one regional market and no institutional scale is evidenced beyond the client count.