Accelex

Thousands of manager specific document formats defeat any rules based parser, which is the whole problem. Proprietary machine learning techniques automate extraction, analysis and sharing of data that is hard to access precisely because it arrives as prose and tables inside fund reports rather than as structured feeds. The extraction reaches down to instrument level across the capital structure, distinguishing redemption priority, interest and principal obligations within private debt. Without it, the work falls back to the expensive and error prone manual processing the company exists to replace.

The platform produces structured data rather than conclusions, and the investment decisions it informs remain entirely with the client, which is the correct division for a data layer. Automation is described as end to end across the workflow from document acquisition through extraction to reporting and analytics, and the acquisition step in particular runs without human involvement by design, consolidating documents from portals, email and file transfer.

What is absent is the review layer that matters most for this product: nothing describes what happens when extraction confidence is low, whether uncertain values are flagged to an analyst before they reach a report, or how a client verifies figures against source documents.

Accuracy is the central gap for this product, not a peripheral one. No accuracy rate, extraction error figure, validation result or reconciliation statistic is published. The company's own framing is that manual processing is expensive and error prone, which asserts a comparison it never quantifies. The output feeds portfolio valuations, performance attribution and client reporting, so an extraction error does not stay in the data layer. It becomes a reported return. What proportion of extracted values are verified, and against what, goes unanswered.

A global custodian bank is named as a customer, and the detail matters: it had already developed its own artificial intelligence extraction capabilities for alternative asset servicing and adopted this platform alongside them to reach fund and underlying asset level look through for its asset owner clients, which is a sophisticated buyer concluding the hard part was worth buying.

The client base is quantified through the assets it represents rather than by logo count, at more than $1.5 trillion invested across over 13,000 private market funds and 4,000 asset managers, with independent profiling recording more than 60 global investors and service providers. The Series A of $15 million was led by a major listed financial data provider whose own private markets strategy already depended on the technology. Six offices span three continents.

The question of a data boundary takes a particular shape for this product, and no boundary statement is published. Extraction models improve markedly with exposure to more manager specific document formats. The platform processes reports from 4,000 asset managers on behalf of more than 60 investors, so what the system learns from one client's documents makes it better at every other client's. The documents add a second dimension.

A fund report is delivered by a general partner to its limited partners under confidentiality, and the extraction platform sits inside that relationship without being party to it. Nothing states what is retained, whether format learning is shared, or how one investor's portfolio is separated from another's.

Subject matter works in its favor. The payload is institutional investment data (capital account statements, fund reports and transaction records) rather than personal information about consumers, so most consumer privacy questions do not arise. What is sensitive is commercial rather than personal, and highly so: a client's complete private markets portfolio, its manager relationships and its returns. There are no public data processing terms, retention schedule or subprocessor list, and the platform operates across six jurisdictions including two European data protection regimes.

There is no public attestation, certification, trust center or enumerated framework. A global custodian bank and a listed financial data provider have both integrated this technology into their own client facing products. That means it has passed vendor security assessment at a demanding standard, and those firms have staked their own client relationships on it. None of the resulting assurance is published.

No supervisor, statute, instrument or reporting standard is named. The omission is notable because private markets reporting has established industry templates and disclosure conventions that determine what limited partners are entitled to receive and in what form, and European fund managers operate under a directive with its own investor reporting obligations. A platform whose purpose is extracting and standardising exactly that reporting operates alongside those frameworks and identifies none of them.

No individual is assessed. The exposure concerns which managers are read accurately. Extraction quality depends on document format, and a large institutional general partner producing standardized quarterly reporting will be parsed far more reliably than a smaller manager, a first time fund or one reporting in another language or convention.

An investor's view of its smaller and less conventional managers is therefore systematically less reliable than its view of its largest, in exactly the part of a portfolio where transparency is hardest to obtain by other means. There is no published coverage or accuracy breakdown by manager type, document format or language.

The exposure runs further than the direct customer, and there is no published guarantee, indemnity or falsifiable accuracy commitment. Extracted values flow into portfolio analytics, performance reporting and, through the custodian and data provider relationships, into products those firms supply to their own clients. An extraction error propagates through several parties before anyone reads it. Nothing describes how an error surfaces, who is notified, or how a corrected figure is reconciled once a report has been issued on the original.

The document acquisition layer is disclosed with specific named sources: two major investor portal providers connected through their interfaces, alongside email capture and file transfer. A buyer knows exactly where the raw material originates and can assess whether its own managers' delivery channels are covered. The extraction models are the company's own and described as proprietary, which shortens the chain at the decisive point. No model provider is disclosed for the underlying language processing, and there is no public subprocessor list or hosting arrangement.

Integration solves the step before extraction, a step competitors often ignore. Fund managers deliver documents through investor portals and email, requiring repeated manual logins and downloads before any processing begins. The platform connects directly to two named major portal providers through their interfaces, alongside email attachment capture and file transfer, to produce a consolidated content feed. Downstream, the technology is embedded in a major financial data provider's private markets offering and integrated into a global custodian's front office platform for its asset owner clients.

Hosting provider, region selection, residency commitment and private deployment options are all unpublished. The company has entities across the United Kingdom, France, Luxembourg, the United States, Canada and Australia. Its clients include custodian banks whose own regulators impose location requirements on outsourced processing. A buyer base like that would ordinarily require a stated residency position.

The charging model is disclosed in structure through independent profiling: revenue comes from annual software subscriptions under standard contracts, with tiers varying by data volume and service level. Those are the two dimensions to plan against. Document volume is the natural cost driver for an extraction platform, and service level determines what verification accompanies it. No rate is published, and the disclosure reaches the reader through a data provider rather than the company.

Six buyer types are served across both sides of the servicing relationship, spanning asset owners, allocators, institutional investors, asset servicers, custodian banks and service providers, so the same platform reaches the investor holding the position and the administrator reporting on it.

Instrument coverage widened deliberately beyond private equity into private debt with instrument level extraction across multiple levels of the capital structure and successive funding rounds, which is where the analytical difficulty concentrates. Geographic presence spans the United Kingdom, France, Luxembourg, the United States, Canada and Australia, the last opened in 2025 to meet Asia Pacific demand for data transparency.