LSEG

The Clearwater band, and this instance sits at the lower end of it. The AI capability is largely supplied by a technology partner and layered onto a licensed data estate: the agent inside the workspace is described by the vendor as a reporting layer built on top of existing financial and market data rather than a new analytical capability. Strip the inference and the data platform, the analytics, the screening content and the market infrastructure all remain untouched. What keeps it above a reject is that the agent is shipped, named and running against the vendor's own content rather than being an announcement or a category essay.

The shipped agent assembles a briefing that an analyst then uses, and the vendor positions the value as shortening the distance to a data backed answer rather than replacing judgement. Held off A for the usual reasons, no published approval gate or scope limit, and for one specific to this architecture that is worth watching as protocol interfaces spread: the governance controls cited in the vendor's own announcement belong to the partner's agent building platform rather than to this vendor. When a customer builds an agent on a third party platform against this vendor's licensed data, whose oversight design applies is left unstated, and a buyer cannot assume it is the data provider's.

The vendor publishes no accuracy rate, validation approach, benchmark or drift policy. Its own executives state that trust and accuracy underpinned by robust data governance are critical in financial services, which identifies the right requirement and discloses nothing about how it is met. Asserting that accuracy matters is not a transparency measure, and this axis credits the second rather than the first.

The vendor names no customer and publishes no quantified outcome for the agent. Two prominent named relationships exist, a technology partner and an indexed artificial intelligence platform, but a partnership is not a customer outcome, a lesson this sweep has already paid for once. The quantified figures published are problem statements about how long pitchbook preparation takes across the industry rather than results the product delivered for anyone.

The vendor publishes no statement on whether customer queries or connected proprietary data are used for training, no retention terms and no isolation position. Responsible and secure use of AI is invoked as a shared commitment in partnership language without any accompanying term a buyer could rely on.

The vendor publishes no processing terms, retention schedule or privacy position for the generative features in the material reviewed. An announced next phase would let firms combine their own proprietary data with the vendor's datasets inside the partner environment, which raises the question directly and is not yet accompanied by any published terms.

The vendor names no certification, audit type or trust portal in the material reviewed. Security is invoked repeatedly as an adjective across the partnership announcements, describing connectivity as secure and seamless, which is language rather than credential. Recorded as an unverified absence, with the trust page check queued alongside the other data majors.

Graded C on the division scoping principle, and this is the sharpest test that principle has faced. The group operates a recognised investment exchange, a clearing house, a trading venue and a benchmark administrator, all supervised, so the parent is about as regulated as a financial firm can be. This record is scoped to the data and analytics division, and none of that authorisation attaches to a research platform or its agent. Holding the line here matters precisely because the temptation is strongest: a reader would otherwise conclude the agent is supervised because the exchange is.

The vendor publishes no evaluation methodology, fairness testing or governance position for the agent's output. The selection question is live here as elsewhere in this pocket: a briefing that assembles peer comparisons, analyst research and deal history is making choices about what to include and what to leave out of a picture of a company, and nothing published describes how those choices are evaluated.

The vendor publishes no liability position, indemnity or remediation route for generated output. The architecture raises a question this index has begun tracking under the agent marketplace liability gap and this is its clearest instance yet: when licensed content is drawn through a protocol server into an agent the customer built on a third party platform, and that agent produces a wrong answer, the responsibility is divided between a data licensor, a platform provider and the customer who assembled the agent. Nothing published by any of them allocates it.

The supplier is named plainly and repeatedly rather than obscured behind a claim of proprietary technology: a single technology partner supplies the AI, its agent building platform and assistant product are named as the deployment path, and the underlying models are acknowledged as that partner's frontier models. For a data business layering inference onto licensed content, that tells a buyer most of what they need to know about who is answering.

Held off the bar set earlier in this sweep because no model family or version is mapped to a specific feature, nothing is administrator selectable, and no failover or alternative provider is disclosed, which leaves the customer with a single undisclosed dependency at the model layer.

The group operates exchange, clearing and post trade infrastructure that institutions transact on, which is the system of record test met at the most fundamental level available in this market, and the data business delivers into customer environments through feeds, analytics and now an interoperability protocol server.

The integration ambition is explicit and unusually far reaching: licensed data activated inside a partner's productivity suite, reaching spreadsheet and messaging applications, and embeddable directly in a client's own platform.

A named cloud partner and a large multi year commitment to it are published, but that is an infrastructure relationship rather than a residency statement, and the distinction is applied here exactly as it was to the other builds in this session. No region, jurisdiction, residency commitment or tenancy position was located. Consequential given a European headquartered group whose licensed data is being routed into a United States technology partner's productivity environment.

The vendor publishes no price, tier or unit of charge for the platform, and independent coverage notes explicitly that it has not said whether the shipped agent is available across all subscription tiers, which leaves a buyer unable to tell whether the capability is included in what they already pay for.

More than 350,000 platform users across over 1,000 applications, spanning the buy side, the sell side, banks, corporates and market infrastructure participants, and global rather than regional in both content and customer base. The data estate is published at roughly 33 petabytes and the merger database alone covers more than 1.5 million transactions. Financial crime screening content extends the reach into compliance functions at institutions that are not otherwise data platform buyers.