FundCount

The Clearwater shape, tested against the Oxane refinement because an extraction led product can fail it. It does not fail here. Ask not whether software remains when the models are stripped but whether the software still has anything to operate on: this platform's primary inputs are automated feeds from named market data providers, custodians and brokers, reconciled by pre built importers, so the general ledger, partnership accounting, allocations, waterfalls, reporting and the investor portal all keep working. The document extraction module is how the alternatives portion of the data arrives, and it is one input path rather than the substrate. C and built.

A human review point is named and positioned, which is more than most extraction vendors manage. Extracted data lands on a dashboard for review by the accounting team or the client in real time before it is used, verification processes are stated to run alongside the automated extraction, and the source document stays one click away from inside any report, so an unusual figure can be checked against the page it came from without leaving the workflow.

That last property is provenance at the point of use. Held off A on the distinction that decides this axis: a dashboard where review is available is not the same as a gate where review is required. Nothing published states a confidence threshold that routes a field to a person, what happens to an extraction nobody reviews, or whether posting to the ledger is blocked until review completes.

The standing question for any document extraction vendor is extraction accuracy, and it is unanswered here. No error rate, no confidence scoring, no human review rate, no benchmark and no drift policy are published. What is published is a capability argument rather than a performance one, and it is a specific and credible argument: the vendor names the failure modes it claims to handle, watermarked files, changing statement formats and unstable element positioning in compound documents, and contrasts that with older optical character recognition.

Naming what you cope with is not the same as saying how often you are right. The consequence is sharper here than in most extraction cases in this index, and worth carrying: this extraction does not feed a report, it feeds a general ledger, so a wrongly read capital account balance propagates into the books and out into an investor's own statement.

No named client, no case study and no quantified outcome located during this pass. Buyer types are described generically and the benefit claims are stated in hours saved and manual work avoided without a baseline or a figure. Review site listings exist and are not creditable under the source test. This is the standard C on this axis: no named customer and self reported framing only.

Nothing published on whether client documents or extracted data inform anything the vendor builds, whether tenants are isolated, or what happens at termination. Notable because the documents in question are other managers' capital account statements and fund financial statements, so a single client's document store contains material about funds that are not that client's own.

No privacy position, processing terms, sub processor list or retention statement located. The material held is investor level: capital account balances, capital call and distribution history and tax schedules for identified individuals and entities.

No certification, attestation or trust portal located in the vendor's own swept material, so nothing is credited. Banked check with a specific reason to expect a result: fund administrators are among this vendor's named buyer types, and an administrator whose own clients demand an attestation will require one from the platform running the ledger, so a security or trust page and a service organisation control report are both likely to exist.

No licence, no supervised test and no programme enrolment, which is the expected position for an accounting software vendor selling to funds rather than to supervised deposit taking institutions.

Nothing published. Consumer protection exposure is among the lowest in this index, since the work is extracting figures from fund documents into a ledger and no decision about a person is being made. Recorded for completeness rather than as a criticism of substance.

No published allocation of responsibility and no route for an affected party. The affected party here is identifiable and is not the customer: an investor whose capital account statement is generated from the same ledger the extraction feeds. If a figure is read wrongly and not caught at the review dashboard, the person who receives the incorrect statement has no relationship with this vendor at all.

The vendor states that extraction runs on a modern large language model and contrasts that class of technology with older machine learning and optical character recognition, which tells a buyer the category of technology and nothing else. No provider, model, version or hosting arrangement is named. The building permits silence pattern holds, with the small variation that here the vendor is explicit about the generation of technology it uses while remaining silent about whose it is.

Named counterparties rather than a category claim, which is what this axis rewards: two major market data providers and four named custodians and brokers feed pricing, positions, foreign exchange and trades automatically, with pre built importers for source specific formats reconciled to the books.

The platform is itself the ledger for its clients, and the extraction module is stated to feed either FundCount or another accounting system, so the output has a published export path rather than a closed loop. Held off A because the integrations are market data and custodial feeds, which are table stakes for an accounting platform, with no partner ecosystem, certified connector list or outward interface published beyond that.

A hosting fee is disclosed as a separate charge, which implies a hosted option alongside something else, and that is the extent of it. No hosting model, region, tenancy or residency statement was located, for a platform that holds the books and the investor records of funds.

Genuinely unusual for this index and earned. The company publishes a starting annual price for its private equity product, a real number with a stated period, and states plainly that transformation and hosting fees apply separately rather than burying the fact that the headline is not the total. Almost nothing in this pocket publishes any number at all.

Held off A because it is a floor price for one product line rather than a rate card, no unit or tier basis is given, and the price of the artificial intelligence extraction module is explicitly not presented in the product materials, which is the part a buyer evaluating this record most wants to know.

Five distinct buyer types published and consistent across sources: hedge funds, private equity firms, family offices, fund administrators and asset managers, sold to fund accountants, controllers and finance officers, with the company describing a global client base.

Held at B rather than A, which is where the larger platforms in this pocket sit, because the segments stop at the fund and the family office: there is no institutional allocator, bank, insurer or asset servicer offering, and no client count or assets figure is published to size any of it.