Endex

The models are the entire product. There is no workflow platform, data asset or system underneath that would survive their removal, because the surrounding environment belongs to Microsoft rather than to this vendor: strip the inference and what remains is an empty spreadsheet add in. The company's own framing makes the point, describing a deliberate decision not to build a competing platform but to place an agent inside the tool finance already uses. Assistant native at its clearest.

Two real oversight properties, both unusual. Every output carries integrated citations attributing it to an underlying source, and workflow tracing maintains labelled cell references so a user can see which cells the agent produced rather than discovering them later. That second one is the more valuable and it is specific to this setting: a spreadsheet is exactly where an unmarked machine edit would be hardest to detect and most damaging.

Held off A because the agent writes into a live financial model and no approval gate before those writes is published, no confidence threshold routes a case to a person, and nothing states what the agent will not do.

Earned on provenance stated consistently rather than asserted once. Outputs carry attribution to underlying sources, conclusions can be traced back through labelled cell references, and the system actively cross checks sources against each other, surfacing discrepancies and flagging restatements disclosed in footnotes with targeted citations. Catching a restatement buried in a footnote and citing it is a checkable behaviour, not a claim about explainability.

Held off A because no accuracy rate, benchmark, evaluation methodology or drift policy is published, and the one accuracy statement located, that a model change simplified the pipeline without sacrificing accuracy, is an assertion with no figure behind it.

The C bar as located. Customer categories are stated but no institution is named, and the testimonials on the vendor's own site are attributed to an unnamed hedge fund portfolio manager. One quoted figure exists, a three statement model and valuation built in ten minutes, but it comes from an anonymous user rather than an identified firm. A model provider has published an account of this vendor's work, which is an independent party naming the relationship, but it evidences the model dependency rather than a customer outcome.

The vendor publishes no statement on whether customer files, models or queries are used to train or improve any model, no retention terms and no tenancy or isolation position. Compliance and security are invoked as commitments without accompanying terms. A buyer can infer where inference happens from the named model provider, which is more than most, but inference location is not a stewardship commitment.

The vendor publishes no processing terms, retention schedule, sub processor list or privacy commitment beyond the adjective already discounted above. The gap is sharper than usual given that the material flowing through the product includes confidential deal documents and a named external model provider sits in the processing path.

The vendor names no certification, audit type, report or trust portal. What stands in their place is a run of adjectives: the product is described as institutional grade and privacy centric, and as maintaining the highest standards of data security and compliance. This index already catalogues institutional grade as an unverifiable quality claim, and privacy centric is the same construction. Third instance in this session of a young vendor substituting a grade adjective for a credential, and it earns nothing.

The vendor holds no licence, registration or supervised status, and none would be expected: it sells a modelling tool to regulated firms rather than performing a licensed activity itself. Worth noting for a buyer that models built with agent assistance may feed valuations and materials that sit inside the customer's own regulatory obligations, and the vendor publishes nothing addressing that boundary.

The vendor publishes no evaluation methodology, testing regime or governance position for the agent's behaviour. The live question in this setting is not protected class fairness but selection and reconciliation: when two sources disagree on a figure, the system decides which to prefer, and nothing published describes how that judgement is made or evaluated even though the reconciliation capability is marketed as a headline feature.

The vendor publishes no liability position, indemnity, accuracy warranty or remediation route. The exposure is unusually direct compared with the research assistants elsewhere in this pocket: this agent writes into a live valuation model rather than producing a document a person reads and judges, so an error enters the arithmetic of a transaction rather than a summary of it.

The most granular supply chain disclosure in this index. The provider is named on the vendor's own site, and a jointly published technical account names specific model versions and maps each to the workflow it serves with a stated reason: one reasoning model replaced a chain of prompts and verification steps for multistep financial reasoning, a smaller model cut latency to roughly a third per turn for long multi step work such as analysing confidential information packages and reconciling models, and several others were evaluated in the process.

A buyer can therefore tell not only whose models are involved but which one answers which kind of question, which is more than any other vendor here discloses. Graded on disclosure rather than architecture: the same transparency reveals a single provider dependency with no alternative or failover described, but that is a concentration the disclosure exposes rather than a gap in the disclosure itself.

Integration depth is the product's central design choice rather than a feature list. The agent operates inside the spreadsheet where the modelling actually happens, unifies a firm's internal files with public disclosures and a named licensed data source into one searchable plane, and delivers into the document and presentation formats the same teams already use. Held at B because the environment it inhabits is another company's, and this vendor is not itself the system of record for any function.

The vendor publishes no hosting model, region, residency commitment or tenancy position. Material here because the product ingests confidential information packages, internal investment files and live models from banks and private equity managers, and a buyer in that position needs to know where that content travels and comes to rest.

The vendor publishes no price on its own site. Flagged as incomplete rather than settled: the product carries a listing on a major software marketplace, and the standing rule here is to check a marketplace listing before grading commercial at C. A first year discount code circulating through a training partner also implies a published list price exists somewhere. Both checks are queued and either could move this grade.

Genuine breadth by institution type across the sell side and buy side: investment banks, private equity managers above 100 billion dollars in assets, global multi strategy hedge funds, corporate finance teams at large listed companies, and listed real estate developers.

Held at B rather than A because every segment is described as a category rather than counted or named, no client number is published, and no geographic coverage is stated, so a reader learns the shape of the customer base and not its size.