AlternativeSoft

The engine underneath is quantitative finance rather than machine learning, and it was the entire product for close to two decades. Four thousand risk adjusted statistics, factor models, and mean variance and risk parity optimization are deterministic computation, as are Monte Carlo simulation and peer group construction. Remove the models and every one of those functions runs unchanged. What the models add is real and confined to two workflows: reading fund documents to pre populate due diligence questionnaires, and generating reports and factsheets.

The oversight statement is explicit and placed at exactly the right point in the workflow. The company describes the questionnaire tool as reading the manager's fund documents and pre populating responses, which the manager's own team then reviews and finalizes. That puts a named human step between a generated answer and a document sent to an institutional investor. What is missing is any description of what the reviewer is shown, or whether generated text is marked as generated. Nothing says how the automated report and factsheet generation is checked before distribution either.

The quantitative side of the platform is transparent by construction, since risk adjusted statistics and optimisation methods are standard published finance rather than proprietary inference. The model side is not measured at all. No accuracy figure exists for reading a fund document and pre populating an answer, no error rate by document type, no validation methodology, no drift monitoring and no revalidation cadence. The single published number measures time saved, which describes speed rather than correctness, and for a tool producing representations relied on by fiduciaries correctness is the whole question.

Six institutional clients are named openly, including a global bank, two large asset managers, a private bank and a European fund manager, alongside a stated hundred and fifty plus institutions managing over $1.5 trillion and a business trading since 2005. The independent element is a risk management software award from the institutional hedge fund industry taken four consecutive years, which is repeated outside evaluation rather than a single placement.

Client voices appear as direct quotes about analytical use rather than generic praise. The one number attached to the artificial intelligence line, a reduction in questionnaire completion time of up to ninety percent, is self reported and not tied to any named institution.

One published sentence makes the stewardship question unavoidable rather than theoretical. Marketing aimed at fund managers offers them a view of where their fund ranks against the peer set allocators screen them in, drawn explicitly from the same database the allocators use.

That establishes a single data foundation serving parties on opposite sides of a selection decision, and nothing published states what does or does not cross between them, whether allocator screening criteria, watchlists or search behaviour inform anything a manager can see, or whether uploaded due diligence material is used beyond the engagement it was supplied for.

Fund performance, manager documents and allocator portfolios make this institutional data rather than consumer financial information, so the statutory frame for consumer data applies weakly. No privacy statement, data handling description or retention terms are published, and nothing explains how manager supplied documents are treated once uploaded. Due diligence files uploaded by fund managers routinely contain named personnel, and nothing addresses that.

A security claim appears in the vendor's own platform specification, and it does not identify a credential. The wording is that the platform is compliant with an international standards body, compliant with European data protection law and carries enterprise grade security. The standards body is named without the standard, so a reader cannot tell which of several hundred documents is meant. Compliant is not the same word as certified, which is the word an audited firm uses.

The third phrase is an adjective. No security or trust page with a numbered standard, an audit report type or a testing statement is public. The platform holds allocator portfolios and manager due diligence files, along with the screening behavior of institutions running over $1.5 trillion, which makes that page the single most valuable thing missing.

As an analytics supplier, the company holds no license and needs none. The regulatory question this product raises sits one step downstream and is unaddressed. Due diligence questionnaire responses are representations made by a fund manager to institutional investors, frequently on standard industry templates. They inform fiduciary allocation decisions and are retained as diligence records.

Nothing published addresses whether a machine generated response should be identified as such to the recipient, what accuracy obligation attaches, or how an allocator relying on the answer would know how it was produced.

The fairness question here is which managers get seen. Screening across five hundred thousand funds on four thousand statistics decides visibility, and quantitative screens systematically favour managers with longer track records, standard reporting formats, larger asset bases and conventional strategies, which disadvantages emerging, smaller and non United States managers regardless of skill.

Nothing is published on how peer groups are constructed, what happens to funds with short or irregular histories, or whether the effects of screen design are examined. The company also sells those same managers a marketplace listing to become visible, which makes the design of the screen commercially consequential in two directions.

The failure mode is consequential and entirely unaddressed. A machine populated questionnaire answer that is wrong becomes a misstatement made by a fund manager to institutional investors, with the manager carrying the consequence rather than the tool. Nothing published states a warranty, a service level or a correction obligation, or allocates responsibility at all.

Nothing states the opposite either: no published disclaimer confirms that the manager remains solely accountable for what the tool produced. The published review step is the only mitigation, and a review step is a process rather than a remedy.

Artificial intelligence runs through the product material as reading documents, completing questionnaires, flagging compliance concerns and generating reports, and no provider, model family, version or hosting arrangement is named for any of it. One published detail invites a mistake worth naming, because a careless reader would count it as a disclosure.

The protocol layer names three widely used assistants a client can connect from, and those are the buyer's own tools rather than the vendor's suppliers. Naming which assistant may reach the data says nothing about whose model reads a manager's offering memorandum and writes the answer that goes to an investment committee.

Nothing distinguishes work performed in house from work called out to an external service, and a fund manager uploading unpublished strategy detail and named personnel cannot establish from public material which counterparties process those files.

The integration estate is enumerated rather than gestured at, and it reaches both ends of the workflow. Fourteen inbound connections are named individually with the content each carries. They cover six fund databases, two private markets sources and a hedge fund research provider, plus a market data terminal and a specialist due diligence data house, all consolidated into one screening universe so the reconciliation work does not fall to the client.

On the outbound side, fund administrators and prime brokers are named as feeding position, exposure and net asset value data over managed file transfer and application interfaces. A spreadsheet connection runs in both directions for every field on the platform, and a live business intelligence connector supports client built dashboards. A published interoperability protocol server supplies a developer facing route, with hosted and local options.

The platform is described as cloud hosted software delivered in a browser with no installation. The protocol layer that exposes fund data to an assistant carries an explicit choice between a hosted server and a local deployment, with computation stated to run against the institution's own data.

A buyer can therefore establish the delivery topology and can keep the assistant facing component inside its own environment if procurement requires it, which is a real decision rather than a marketing line. Data residency remains the unanswered half. No hosting regions or storage location for uploaded manager documents are published, and there is no stated single tenancy or contractual location commitment. The buyer base includes sovereign wealth funds and pension plans, the institutions most likely to carry a location requirement.

No rates are published, but the billing basis is. The company states that pricing is module based and that institutions pay for the specific modules they need. The figure varies with institution size, module selection and the number of connected data provider feeds. A buyer can therefore see which levers move the price and can reason about whether a narrow deployment is cheaper than a broad one before entering a sales process.

The buyer list is broad and enumerated: pension funds, endowments, foundations, sovereign wealth funds, family offices, funds of funds, private banks, wealth managers and financial advisers on the allocator side, with more than a hundred and fifty institutions and over $1.5 trillion stated. Asset class coverage spans hedge funds, private equity, real assets and multi asset strategies inside one analytical environment rather than through separate products. The company also sells to fund managers, giving it both sides of the allocation relationship, and the named client base spans the United Kingdom, continental Europe and the United States.