Endex

The models are the entire product. No workflow platform, data asset or system lies underneath that would survive their removal, because the surrounding environment belongs to Microsoft. Strip the inference and what remains is an empty spreadsheet add in. The company's own framing makes the point: it describes a deliberate decision not to build a competing platform but to place an agent inside the tool finance already uses.

Two real oversight properties are built in. Every output carries integrated citations attributing it to an underlying source. Workflow tracing maintains labeled cell references, so a user can see which cells the agent produced rather than discovering them later. The second 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.

The gap is that 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.

Provenance is stated consistently rather than asserted once. Outputs carry attribution to underlying sources, and conclusions can be traced back through labeled cell references. 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 behavior, not a claim about explainability.

No accuracy rate, benchmark, evaluation methodology or drift policy is published. The one accuracy statement, that a model change simplified the pipeline without sacrificing accuracy, is an assertion with no figure behind it.

Customer categories are stated, but no institution is named. The testimonials on Endex'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 the company's work. That is an independent party naming the relationship, but it evidences the model dependency rather than a customer outcome.

Whether customer files, models or queries are used to train or improve any model is not stated. There are no published retention terms and no tenancy or isolation position. Compliance and security are invoked as commitments without accompanying terms. The named model provider shows where inference happens, but inference location is not a stewardship commitment.

Beyond the privacy centric label, no processing terms, retention schedule, subprocessor list or privacy commitment is published. The gap is sharp because the material flowing through the product includes confidential deal documents, and a named external model provider is in the processing path.

Adjectives stand in for credentials. The product is described as institutional grade and privacy centric, and as maintaining the highest standards of data security and compliance. Institutional grade is an unverifiable quality claim, and privacy centric is the same construction. No certification, audit type, report or trust portal is named.

No license, registration or supervised status is held, and none would be expected. Endex sells a modeling tool to regulated firms rather than performing a licensed activity itself. Models built with agent assistance may still feed valuations and materials that sit inside the customer's own regulatory obligations, and nothing published addresses 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.

An error here enters the arithmetic of a transaction, not a summary of it. Research assistants produce a document a person reads and judges. This agent writes into a live valuation model instead. No liability position, indemnity, accuracy warranty or remediation route is published.

Supply chain disclosure is granular. The provider is named on Endex'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 multistep work, such as analyzing confidential information packages and reconciling models.

Several others were evaluated in the process. A reader can tell not only whose models are involved but which one answers which kind of question. The same transparency reveals a single provider dependency with no alternative or failover described, a concentration the disclosure exposes rather than a gap in it.

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. The environment it inhabits is another company's, and this vendor is not itself the system of record for any function.

Nothing published says where customer content travels or comes to rest. That matters because the product ingests confidential information packages, internal investment files and live models from banks and private equity managers. No hosting model, region, residency commitment or tenancy position is stated.

Endex publishes no price on its own site. Two signals suggest a price exists elsewhere. The product carries a listing on a major software marketplace, and a first year discount code circulating through a training partner implies a published list price. Neither source is reflected here.

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