MSCI

The Clearwater band with a nuance this index has recorded before and should keep separate: the underlying estate is quantitative finance modelling, factor and risk models, which are the wrong kind of model for this axis rather than an inference capability. What earns the build is the layer above them, a shipped generative product with natural language interaction, narrative generation and anomaly detection, sold under its own name. Strip that layer and the risk analytics business is entirely intact, which places it at C rather than higher.

The product surfaces, summarises and flags, and a risk manager acts. Anomaly detection and limits monitoring raise items for attention rather than enforcing an outcome, and the generated narrative is explicitly framed as helping identify the most salient information in a risk report rather than replacing the report. Held off A because no approval gate, confidence threshold or escalation rule is published, and nothing states what the generated summary should not be relied on for.

No accuracy rate, validation approach, benchmark or drift policy is published for the generative layer. The contrast is the sharpest available anywhere in this index: this firm's factor and risk model methodology documentation is an industry reference, used by clients to satisfy their own model validation obligations, and none of that discipline is visible for the generated narrative sitting on top of those models.

A user receives a written interpretation of a model output with no statement of how that interpretation is evaluated. The methodology publication itself was not verified in this pass and the check is queued.

The vendor names no client using this product and publishes no quantified outcome for it. Claims are capability descriptions and awards language about the underlying analytics rather than evidence that the generative layer produced a result for anyone. Recorded as an unverified absence rather than an evidenced one, with the customer story check queued.

The vendor publishes no statement on whether client portfolio data is used to train or improve models, no retention terms, and no tenancy or isolation position. Pointed here because the product ingests a client's actual portfolio positions to generate its commentary, and position level holdings are among the most commercially sensitive data an asset manager holds.

The vendor publishes no processing terms, retention schedule or privacy position for the generative product in the material reviewed. The data at issue is institutional portfolio holdings rather than consumer information, which changes the character of the exposure but not the absence of disclosure.

The vendor names no certification, audit type or trust portal in the material reviewed. Recorded as an unverified absence: no trust or security page was reached in this pass, and the check is queued alongside the other data majors built this session.

Graded C on the division scoping principle, the fifth application across recent sessions. Group subsidiaries administer indexes and are authorised and supervised as benchmark administrators in the United Kingdom and the European Union, which is a genuine authorisation, but this record is scoped to the risk analytics business and that supervision attaches to index administration rather than to analytics software. Flagged as a live check rather than settled: if generative or analytical output feeds into an administered benchmark, read across becomes arguable and would move the grade.

The vendor publishes no evaluation methodology, fairness testing or governance position for the generative layer. The live question here is not protected class discrimination but selection: the product decides which changes in a portfolio are salient enough to surface before the working day starts, and a systematic tilt in what an anomaly detector raises or omits shapes what a risk manager sees. Nothing published addresses how that selection is evaluated.

Graded C, but for a reason that differs from every other C on this axis and is worth recording. Most of this index is silent on liability. This vendor is the opposite: the product's own factsheet carries an explicit and comprehensive disclaimer stating the information is provided as is, that the user assumes the entire risk of any use made of it, and that neither the company nor its subsidiaries nor any information provider makes any warranties.

That is a clearly published liability position, and it allocates all of it to the customer. It earns no credit here because this axis measures recourse rather than disclosure of its absence, and the two grades above C in this index were both given for actual mechanisms an affected party could use. A buyer is nonetheless better informed by this than by silence, which is a distinction the axis currently cannot express.

No model provider, family or version named for the generative layer. Third instance in this session of a distinction now worth treating as settled: a named cloud host, a named data warehouse and a published interoperability protocol are all infrastructure and access disclosures, not supplier disclosures. Naming where inference runs or how results are reached says nothing about whose model produces the words.

A strong integration story stated concretely rather than in generalities: the vendor runs managed extraction and loading, delivers risk and performance statistics into a cloud data warehouse, and exposes results through warehouse integrated data models, a named connector, curated dashboards, direct warehouse access and a client's own assistant environments. That is analytics travelling into the customer's stack on the customer's terms. Held at B because it delivers into the systems of record rather than being one.

The delivery architecture is described in useful detail, including managed loading into a cloud data warehouse and a named warehouse technology for access, but that is a product description rather than a deployment and residency statement. No region, jurisdiction, residency commitment or tenancy position is published, and a buyer cannot learn where portfolio holdings come to rest. The distinction is the same one that separates a corporate footprint from a residency statement and it should be applied consistently.

The vendor publishes no price, tier or unit of charge on its own pages. Flagged as incomplete rather than settled: the product carries a third party software marketplace listing, and the standing rule in this index is to check a marketplace listing before grading commercial at C, because listings often carry contract pricing and structured fields vendors never put on their own sites. That check was not run and is queued.

The generative product is addressed to a defined institutional set: risk and portfolio managers at asset managers, hedge funds and asset owners. That is genuine buy side breadth and global in reach, but it is narrower by institution type than the other data majors in this pocket, with no banking, insurance or corporate segment addressed by this product line. The wider company reaches further through indexes and research, and that breadth is not credited here because this record is scoped to the analytics business.