Dasseti

Models carry meaningful throughput without being the whole product. Extraction from portable documents, word processor files and adviser registration filings, automated flagging and scoring of responses, and generation of committee ready reports from structured data are all doing work a person would otherwise do at volume. But the company sold digitised questionnaires, workflow, relationship records and analytics before any of it, and strip the inference and that platform still functions.

Held at B rather than the C given to legacy platforms because the inference is described as embedded across the whole workflow rather than added at one point, and rather than A because the questionnaire and workflow spine is genuinely independent of it.

A verification step is described rather than implied: responses are pre filled and then verified, which places a human between generation and submission, and flagged risks are raised for a team to assess rather than acted on. Generated reports go to an investment committee, which is a human decision point by construction.

Held off A because no approval gate is published as a control, no confidence threshold routes an uncertain extraction to a person, and nothing states what the embedded assistant is prevented from doing.

The vendor publishes no accuracy rate, validation approach, benchmark or drift policy for extraction, flagging or generation. Accuracy and context awareness are asserted as qualities. The nearest genuine property is that generated reports are described as drawn directly from structured responses and source documents, which points toward provenance, but it is stated as a description of how the writing tool works rather than as a guarantee that every generated assertion traces to a record, which is the distinction this axis turns on.

Named clients with published case studies, which is more than most of this pocket manages: a large insurer using the platform to monitor a multi asset portfolio, an institutional manager streamlining manager oversight, and a named outsourced due diligence provider quoted on why it selected the platform. Held off A on the shape this index sees repeatedly: the named reference and the number never join.

The quantified claims, roughly fivefold productivity improvement and a quarter of work compressed into three weeks, are attributed to due diligence teams generally rather than to any of the named clients.

A specific and unusually useful commitment: the platform is stated to call a named cloud hosted model service configured for zero data retention, so a buyer learns both where inference happens and that the material sent there is not kept. Most vendors in this index answer neither question.

Held off A because the commitment covers the external model service rather than the platform itself, no statement addresses whether customer content is used to train or improve anything the vendor controls, and no tenancy or isolation position is published for the questionnaire and document estate.

The vendor publishes no processing terms, sub processor list, retention schedule or privacy commitment for the platform as a whole. The zero retention commitment located applies to the external model service specifically and is credited under stewardship rather than here, so the same disclosure is not counted twice.

A named certification stated as held rather than aligned with, an annual style attestation covering the platform, alongside a statement that the system has been penetration tested and passed review by large institutional clients. That combination of a specific credential plus independent technical testing is above the norm for a vendor this size. Held off A because only one certification is named, no trust portal or report request route was located, and the approval claim was reached in truncated form so its full scope is unverified.

The vendor holds no licence, registration or supervised status and none would be expected: it supplies software to regulated allocators and managers rather than performing a licensed activity. Worth noting that the product is positioned partly around helping customers meet their own obligations, including flagging regulatory gaps in manager responses against named directives and filing regimes, which is compliance capability rather than the vendor's own standing.

The vendor publishes no evaluation methodology, testing regime or governance position for the inference layer. The live question here is selection and scoring rather than protected class fairness: the system flags which manager responses look risky and scores them for comparison, so a systematic tilt in what gets flagged shapes which managers receive scrutiny and which pass quietly. Nothing published describes how that flagging is evaluated.

The vendor publishes no liability position, indemnity, accuracy warranty or remediation route. The consequence is specific in this product: a missed disclosure or a wrongly cleared response feeds an allocation decision on behalf of pension and endowment beneficiaries, and the audit trail the platform maintains would record what was reviewed without allocating responsibility for what the model failed to surface.

The model service is named plainly on the product page, which for a vendor of this size is better than most of the index manages, and it is named as part of a security and privacy statement rather than buried in a partnership announcement.

Held off A because no model family or version is identified, none is mapped to a specific feature, nothing is administrator selectable and no alternative or failover is described, so a buyer knows the provider but not which model produces a flag, an extraction or a report.

Meets the system of record test for the function it serves. For a due diligence team this is where the manager record lives, where questionnaires are issued and tracked, where documents and meeting notes are centralised and where the audit trail a reviewer or investigator would ask for is maintained.

Named integrations extend it outward: a direct connection into a major consultant database, so database narratives can be managed alongside questionnaires in one workflow, and a partnership with an investment management platform for monitoring operations.

The platform is described as cloud hosted and the model service it calls is named, which tells a reader something about the processing path, but no region, jurisdiction, residency commitment or tenancy position is published. Material for a customer base spanning European, North American and African institutions whose manager files include confidential fund documentation.

The vendor publishes no price, tier or unit of charge, and every commercial route terminates in a demo request. Nothing states whether the embedded inference features carry an incremental charge or sit inside the base platform.

Four distinct institution types addressed with separate products rather than one audience with a second page: asset owners, investment consultants, asset managers and general partners, plus fund of funds teams and outsourced due diligence providers. Clients are published as holding more than 20 trillion dollars in combined assets. Reach is genuinely global, with named clients and partners in Europe, North America and southern Africa and a European private capital association partnership. The sustainability module covers reporting frameworks spanning several jurisdictions, which widens the regulatory footprint further.