Bridgewise

The removal test leaves nothing, since the product is analysis that no human research desk could produce at this breadth. The architecture is stated in two parts: a machine learning model trained on more than twenty years of historical data that scores over 50,000 listed securities, and a proprietary micro language model that renders the resulting analysis as readable reports in the user's own language.

A conversational product answers investment questions across equities, funds, crypto, indices and macro events, and the company acquired technology converting news, social media and regulatory filings into machine readable sentiment data.

This carries the highest stakes of any autonomy position in the index and the least description. The system issues buy and sell recommendations directly to 35 million end users through their brokerage and exchange interfaces, with a robo advisory product making personal recommendations against actual portfolios, and no human review, escalation path, confidence threshold or suitability checkpoint appears anywhere between the model and the retail investor. The company's licensed status establishes accountability for the advice, which is meaningful, and it is a legal position rather than an operational control.

The company makes explainability and auditability its central argument rather than an afterthought, contending that general purpose language models lack the domain rigour and governance regulated investment environments require, and its architecture supports that: a constrained domain specific model operating over structured data is materially less prone to fabrication than a general model, and it is trained on more than twenty years of history.

Held at B for a specific reason: the claim that responses are guaranteed free of fabrication is stronger than any system can support, and publishing it as a guarantee without an error rate, evaluation method or independent validation invites exactly the scrutiny the company says institutions are applying.

Four national stock exchanges are named partners, which is among the hardest institutional customer sets in finance to win, alongside a major market data provider, two large retail brokerages and a leading trading platform. Adoption is stated and growing sharply, from over 50 institutional clients and 25 million end users in February 2026 to over 100 institutional clients and 35 million end users by July, with around 6,000 analysts using it for company research at an earlier point.

It operates from eight offices across five continents, raised a 21 million dollar Series A, made an acquisition, and its advisory board includes a former national exchange chief executive and former research heads at established brokerages.

No boundary statement was located. The platform sits inside competing brokers and exchanges simultaneously, observing what tens of millions of investors ask about and which securities draw attention, which is commercially valuable information about order flow interest. Nothing states whether query patterns or portfolio data from one institution's users inform anything served to another, or what an institution's users contribute by using the embedded product.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The robo advisory product makes personal recommendations based on users' existing portfolios, which means holdings data for individual investors across many jurisdictions passes through the platform, and the conversational product records what those investors ask about. Neither the handling nor the retention of that material is described.

No attestation, certification, trust centre or enumerated framework was located. National exchanges and major brokerages have completed supplier assessment before embedding the platform into customer facing systems, which is among the more demanding reviews in financial services, and none of the resulting control documentation is published for other institutions to rely on.

The company holds a financial adviser licence and its chief executive states the position plainly, that it operates no proprietary fund and is licensed as an adviser, which is what makes issuing actual buy and sell recommendations lawful rather than merely commentary, and simultaneously answers the operator question that disqualifies several similar businesses. Recommendations are described throughout as regulatory compliant.

Held at B because neither the licensing regulator nor the jurisdiction is named, and the company distributes recommendations across more than fifteen markets with materially different rules on investment advice, suitability and disclosure.

The access argument is real and a customer states it better than the company does: a national exchange with around 4,000 listed stocks observes that providing detail on each individual stock was never its area of expertise, and that the platform fills information gaps in areas not covered by equity analysts, in multiple languages.

Most listed companies worldwide carry little or no analyst coverage, and coverage that exists is concentrated in large caps and in English, so automated multilingual analysis reaches both neglected securities and investors excluded by language.

Held at B because the acquired sentiment capability draws on news and social media, where coverage intensity itself reflects size and prominence, so the underlying attention bias can re-enter through that channel, and no fairness or outcome testing is published.

No guarantee, indemnity or correction process was located, and the exposure is unusually direct. Thirty five million retail investors receive buy and sell recommendations generated by the system, and nothing describes what happens when one is wrong: whether the investor is told the recommendation was machine generated, whether the reasoning can be inspected, who bears responsibility between the vendor and the distributing broker, or how a complaint is handled. The stated guarantee against fabricated responses is offered as reassurance rather than as a commitment with any remedy attached.

The core model is proprietary and named as a micro language model purpose built for capital markets conversations, which tells a buyer the dependency is internal rather than resting on a general purpose interface, and that distinction is the company's central commercial argument. The sentiment capability was acquired outright rather than licensed, and one external data partner is named for exchange traded fund coverage. Held at B because the fundamental market data underpinning the scoring model has no named source, and no subprocessor list or hosting arrangement appears.

Delivery is designed for embedding rather than for a destination site, offered as plug and play widgets or interface integration so a broker's own platform carries the analysis, with the company noting setup can run through the client's existing technology partner. Distribution reaches inside exchanges, brokerages and trading platforms, and a named data partnership adds several thousand exchange traded funds to fund coverage. Held at B because no core brokerage, order management or portfolio system is named and no developer documentation was located.

No hosting provider, region selection, residency commitment or private deployment option was located. The company operates from eight countries and embeds inside national exchanges and regulated brokerages, several in jurisdictions with explicit requirements about where investor data may be processed, and published material does not address any of it.

No pricing, packaging or basis of charge was located across a product set spanning reports, conversational chat, fund analysis, bond analysis and embeddable widgets. With customers ranging from a national exchange to a retail brokerage serving millions, whether charge follows end users, securities covered or institutional licence is the question that determines who can adopt it, and none of it is published.

Buyer coverage spans exchanges, banks, trading platforms, investment houses, wealth advisers and financial media and education platforms, which is nearly every institutional participant that touches retail investment. Asset coverage runs across equities on more than 130 exchanges, funds and exchange traded products, bonds, crypto, foreign exchange and commodities, reported at over 70,000 instruments and more than 90 percent of listed global securities. Language coverage of fifteen or more, with offices across Asia, Europe, the Middle East and the Americas, makes the geographic claim credible rather than aspirational.