Arch

The company started in 2018 as automation for collecting partnership tax schedules, and the shape of the product still reflects that origin. Strip the models and a working platform remains: portal login automation, document retrieval, capital call workflow, payment routing and reporting are engineering and integration work rather than model work, and they were the business for several years before any model shipped.

What the models do now is genuine and additive, reading account statements, investor letters and fund updates and summarising the financial data and commentary inside them, which is the step that previously required a person. Models sit on top of a product that would still function without them.

The system acts rather than only reports, which raises the bar, and at the point where it acts on money the company does publish a control. Capital call workflow automation carries built in fraud detection, and that is a well chosen defence, because fraudulent capital call notices sent to family offices are one of the recognised attack patterns in private markets and the loss is immediate and rarely recoverable.

What is not described is the approval path around it: nothing published states whether a person authorises a payment before it moves, what threshold triggers review, or what happens to a call the detection flags. Automation of collection and summarisation carries no described review step either.

Searched for an extraction accuracy figure, an error rate by document type, a description of validation or reconciliation controls, drift monitoring, revalidation cadence or model documentation, and located none. The published material describes what the models do and the volume they handle without any measure of how often they are right.

The fraud detection component is the sharpest instance of the gap: a detection model has a false negative rate, that rate is the whole question for a control standing between a family office and a fraudulent wire, and no figure or testing account is published.

The scale claims are specific enough to be checkable and the buying evidence is unusually direct. More than two hundred and fifty billion dollars of private market assets sit on the platform across more than four hundred and fifty clients, up from a hundred billion fourteen months earlier, and the client base is broken out by type rather than aggregated. The decisive item is not a testimonial.

A major global accounting firm adopted the platform into its own family office service and, according to third party reporting, moved off a direct competitor to do so, which is an independent party staking its own client relationships on the choice. A named chief financial officer of a family office also sits on the client advisory board and speaks publicly about the implementation.

Searched for any statement on where client data goes, whether documents inform models serving other clients, whether the platform builds an aggregate dataset from what it collects, and whether participation in anything of that kind is optional, and located none.

The absence is worth naming rather than passing over, because a platform sitting across four hundred and fifty allocators holds a view of private markets flows that would be commercially valuable if aggregated, and the families and advisers supplying that material have no published assurance about it either way. Nothing suggests misuse, and nothing establishes a boundary.

This is the first vendor in the private markets group where consumer financial privacy law genuinely bites rather than sitting off frame. The platform holds partnership tax schedules, capital account statements and complete private holdings for individual families and the wealth advisers serving them, which is personal financial information about identifiable people obtained through a financial institution relationship.

Searched the platform, security and company material for a privacy statement covering that data, retention, deletion on termination or the treatment of household level information, and located none beyond a general description of secure aggregation. The category shift raises the stakes and the disclosure has not followed it.

Searched the platform, company and press material for an enumerated attestation, an information security standard, a penetration testing statement, a trust centre or a security page and located none, with security described only as secure aggregation in a protected environment.

For a vendor that stores credentials to log into third party portals on behalf of clients, holds tax documents for wealthy families, and initiates payments, this is the widest exposure in the profile, and the standing note applies: asserting security without enumerating it invites an assumption the buyer cannot check.

No licence is claimed and the company presents itself as software, which for the reporting side of the product is the correct posture and is not penalised here. The payments side raises a question the published material leaves open.

A product that automates capital call payments sits close to money transmission, and nothing located states whether funds are held or moved by the company, whether a sponsoring bank or licensed partner stands behind the flow, or which registrations if any apply. Buyers include banks and registered advisers who will have asked that question in diligence, and the answer is not public.

The fairness frame that fits here is accuracy distribution rather than protected classes. Extraction and summarisation quality track document format, and format tracks the size and sophistication of the fund administrator or manager producing the statement, so the smallest and least institutional managers in a family's portfolio are the ones most likely to be read wrongly. Those are frequently the positions with the least independent oversight to catch the error. No accuracy breakdown by document type, administrator or manager size is published, and a summarised fund update reaches an individual investor with no indication of which parts were machine read.

The loss path here is shorter than anywhere else in this lane because money moves. A fraudulent capital call that passes detection produces an immediate and generally unrecoverable wire loss, and a misread statement produces a wrong figure in reporting delivered to a family. Nothing published states a service level, a warranty, a correction obligation, an indemnity or any allocation of loss between the platform and the client for either failure mode. The presence of a fraud detection control alongside no published statement of what happens when it misses is the specific gap, since offering the control invites reliance on it.

The published material describes artificial intelligence summarising financial data and commentary from statements and portals without naming a single model, provider or hosting arrangement behind it. Nothing distinguishes work done in house from work called out to an external service, and nothing states whether the tax schedules and capital account statements of identifiable families pass to a third party during processing.

Two competitors in this lane answer that question directly, one by naming what it refuses to use and one by publishing the connector it exposes, so the disclosure is demonstrably available rather than commercially impossible.

Integration runs in both directions and two named partners sit on the outbound side, an established family office accounting and reporting platform and a global accounting firm's family office data service, each taking private markets data into a system the client already runs for the rest of its wealth picture. Inbound, the whole product depends on reaching fund administrator and general partner portals, which is real and difficult integration work.

What holds this below the lane leaders is the absence of published numbers or names on that inbound estate, with no count of connected portals or administrators and no developer facing interface described.

Delivery is vendor hosted cloud software and the published material says nothing further. Searched for hosting regions, residency commitments, a single tenant option or any account of where documents are stored and processed, and located none.

The client base is described as global and includes four of the twenty largest banks, institutions that routinely impose location and segregation terms by contract, so a buyer has to reach the answer through procurement rather than through anything the company publishes.

No rates, tiers or unit of charge appear anywhere and the route to a number is a demo request. The gap matters more than usual because the client base runs from a single family office to one of the twenty largest banks in the world, a range across which any sensible price must vary by orders of magnitude, and because a payments product sits alongside the software with nothing published on whether it is priced separately, per transaction, or bundled.

Coverage is enumerated with more precision than almost anything else in this index. The published base names single family offices including several with portfolios above thirty billion dollars, registered investment advisers and multi family offices, four of the twenty largest global banks, seven of the twenty largest accounting firms, fund administrators, law firms and institutional allocators, at more than four hundred and fifty clients in total.

That spread covers the adviser, the bank wealth desk, the professional services firm and the institution at once, which is a wider set of buyer types than the private markets data peers reach, and it is supported by named integrations into two established family office reporting platforms.