FundCount

The platform still works without its models. An extraction led product can depend on extraction for everything it does, so the useful question is whether the software still has anything to operate on when the models are stripped. Here it does. Its primary inputs are automated feeds from named market data providers, custodians and brokers, reconciled by pre built importers.

The general ledger, partnership accounting and allocations keep working, and so do waterfalls, reporting and the investor portal. The document extraction module is how the alternatives portion of the data arrives, one input path rather than the substrate.

A human review point is named and positioned. Extracted data lands on a dashboard for review by the accounting team or the client in real time before it is used, and the company says verification processes run alongside the automated extraction. 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 is provenance at the point of use.

Still, a dashboard where review is available is not the same as a gate where review is required. Nothing published sets a confidence threshold that routes a field to a person, or says what happens to an extraction nobody reviews. Nor is it clear whether posting to the general 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, confidence scoring or human review rate is published, and no benchmark or drift policy. 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 than for extraction that feeds a report: this extraction feeds a general ledger, so a wrongly read capital account balance propagates into the books and out into an investor's own statement.

Benefit claims come in hours saved and manual work avoided, with no baseline or figure. No named client, case study or quantified outcome is published, and buyer types are described generically. Review site listings exist, but they do not verify outcomes.

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.

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

Fund administrators are among the named buyer types, which gives a specific reason to expect security credentials, yet no certification, attestation or trust portal is published. An administrator whose own clients demand an attestation will require one from the platform running the general ledger, so a security or trust page and a SOC 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.

Consumer protection exposure is low, and nothing is published on governance or bias. The work is extracting figures from fund documents into a ledger, and no decision about a person is being made, so the gap is a matter of completeness rather than 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.

Extraction runs on a modern large language model, the vendor says, and it contrasts that class of technology with older machine learning and optical character recognition. That tells a buyer the generation of technology in use and nothing else. No provider, model, version or hosting arrangement is named, so whose model it is stays unsaid.

The integrations name counterparties rather than a category. Two major market data providers and four named custodians and brokers feed pricing and positions, plus foreign exchange and trades, automatically, with pre built importers for source specific formats reconciled to the books. The platform is itself the general ledger for its clients, and the extraction module feeds either FundCount or another accounting system, so the output has a published export path rather than a closed loop.

Those integrations are market data and custodial feeds, which are table stakes for an accounting platform. No partner ecosystem, certified connector list or outward interface is published beyond them.

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 is published for a platform that holds the books and the investor records of funds.

FundCount publishes a starting annual price for its private equity product, a real number with a stated period. It states plainly that transformation and hosting fees apply separately rather than burying the fact that the headline is not the total. The number has limits. It is a floor price for one product line rather than a rate card, and no unit or tier basis is given. The price of the AI extraction module is explicitly not presented in the product materials, and that is the part a buyer most wants to know.

Five distinct buyer types are published and consistent across sources. Hedge funds, private equity firms and family offices are named, along with fund administrators and asset managers, sold to fund accountants, controllers and finance officers, and the company describes a global client base. The segments stop at the fund and the family office. There is no offering for institutional allocators or banks, nor for insurers or asset servicers, and no client count or assets figure is published to size any of it.