S&P Global Market Intelligence

The Clearwater precedent, consistent with the two peers already built in this pocket. A data and analytics business of enormous scale with a genuine and broad shipped inference line layered over it, spanning an assistant, document analysis, chart explanation, sentiment scoring, transcript summarisation, natural language screening and generative newsletters.

Strip the models and the platform remains fully usable: the underlying asset is the data estate and the workflows around it, not the inference. It is not the rejected shape either, since the AI is sold as named products across several workflows rather than sitting around the edges of one.

The assistant answers questions and the analyst decides, and the boundary is reinforced by design rather than asserted: answers are built around click through to the underlying source document, so a user is expected to verify rather than accept. Nothing in the line executes a transaction or commits an action. Held off A because no approval gate, confidence threshold or escalation rule is published, and nothing states what the assistant is prevented from doing or where it should not be relied on.

Earned on a provenance property stated consistently rather than on an adjective, which is the distinction this axis turns on. The assistant is described as providing full traceability in its answers, and this is presented as an extension of the platform's long standing principle of click through to source data, so a generated assertion can be followed back to the filing, transcript or record that produced it.

That is checkable in a demonstration, and the claim appears consistently across the vendor's own product page and successive product announcements rather than in one marketing phrase. Held off A because no accuracy rate, validation approach, benchmark, evaluation method or drift policy is published for any generative feature.

The C bar as located: no named customer with a quantified outcome was found in this pass, and the figures published describe the size of the data estate rather than a result the product produced for anyone. Executives are quoted, but in the company's own announcements. Marked explicitly as an unverified absence rather than an evidenced one: a case studies library exists on the division's own site and was not opened. That named check is queued and, on the record of this rule elsewhere in the sweep, it is more likely than not to move this grade.

The assistant is stated to be trained on the platform's own tabular and textual corpus, which is a clearer training disclosure than most vendors offer and points away from customer data as the source. It stops short of a commitment: the vendor publishes no statement on whether customer queries, uploads or usage feed model improvement, no retention terms, and no tenancy or isolation position. A page on the division's own AI ready data exists and was not opened, so the check is queued.

The division publishes no processing terms, retention schedule or privacy commitment specific to the inference products in the material reviewed. Recorded as an unverified absence on the same basis as the security and deployment rows, with the check queued.

The vendor names no certification, audit type or trust portal in the pages reviewed. Recorded as an unverified absence rather than an evidenced one, and flagged prominently: no trust or security page was reached in this pass, so a C here reflects where the research stopped as much as what the vendor publishes. Queued check, and it should be run before this record is used in any comparison against a peer whose trust centre has been read.

Graded C on the division scoping principle, now applied for the third time in two sessions and settled. The corporate parent contains a registered and supervised credit rating agency subject to methodology disclosure obligations, but this record is scoped to the data, analytics and enterprise software division, and that supervision does not read across to unregulated software. The company draws the same division boundary itself throughout its own navigation.

The vendor publishes no fairness testing, evaluation methodology or governance position for its generative features in the material reviewed. The exposure differs from consumer lending because the credit scoring products here assess companies rather than individuals, so protected class doctrine does not apply directly, but systematic slant in which sources an assistant surfaces and which it omits when summarising a filing or a transcript is a governance question the vendor addresses nowhere.

The vendor publishes no liability position, indemnity, accuracy warranty or remediation route for generated output. Consequential given the audience: summaries, sentiment scores and generated screening criteria feed investment banking, asset management and private equity decisions, and nothing published addresses what follows when a generated summary misstates what a filing says.

No external model provider, family or version is named in the material reviewed. The assistant is attributed to the group's own in house artificial intelligence centre, which identifies who built it but not whose models run underneath it, and a subsidiary is not a supplier disclosure. Recorded as an unverified absence: the division publishes a dedicated page for partners and developers that was not opened, and it is the obvious place for supplier naming to appear. Queued, and no comparison should be drawn against a peer that names its model partners until it has been read.

Passes the system of record test on the enterprise services side rather than through the research platform. The division operates securities processing, corporate actions, loan portfolio management and trade and transaction reporting for institutions, which is performing an operational function rather than supplying data into someone else's system.

On the platform side, integration runs deep in the other direction: structured data delivery and feeds, cross reference services that map entities into a customer's own databases, a data marketplace, and office tools that push data directly into customer spreadsheet and presentation models.

The division publishes no hosting model, region, residency commitment or sub processor list in the material reviewed. Recorded as an unverified absence: this vendor almost certainly maintains enterprise trust material that was not reached in this pass, and the check is queued.

The platform publishes no price, tier or unit of charge. Every commercial route terminates in a demo request, a brochure download or a follow up form, and independent comparison coverage of this category describes the pricing as enterprise quoted per seat and unpublished.

Among the broadest in the index. Financial institution segments are addressed separately rather than listed: commercial banking, insurance, investment banking, investment management, and private equity and venture capital, alongside corporates, government and regulatory agencies, defence, professional services and academia. Coverage is global, with country risk research spanning 236 countries and territories and estimates drawn from more than 110 countries. Published scale runs to 109,000 public companies, 60 million private companies and 110,000 private equity and venture funds.