DataVisor

The removal test leaves rules and labelled models, which is exactly the approach the company was founded to displace. Its patented unsupervised machine learning analyses unlabelled data and discovers correlations on its own without being told what categories to look for, which is the core capability, and three further layers sit alongside it: supervised models, graph based link analysis and agents that automate investigation work and rule tuning. Detecting a coordinated fraud ring nobody has seen before is not achievable by any other means.

The governance disclosure attached to the agent layer is the most complete in this index and sets the benchmark others should be measured against. Four controls are named together: interactions are logged, actions require human approval, the whole is auditable, and actions can be rolled back. Rollback in particular appears nowhere else here, and it matters because it is the difference between a system a fraud team can trust with execution and one they must supervise keystroke by keystroke.

An independent practice director frames the effect correctly, that combining conversational and execution agents puts more control in the hands of financial crime leaders rather than less. Real time blocking decisions still sit inside institutional detection thresholds that are not described.

The architecture is disclosed in detail and each of its four layers answers a specific failure of the others: unsupervised learning catches attacks with no labelled history, supervised models apply known signatures efficiently, graph analysis exposes coordination invisible at the transaction level, and agents automate rule tuning so configurations do not decay.

The company states the rationale plainly, that institutions should not rely solely on historical labels or rigid rules, which is the correct diagnosis of why fraud systems fail. Independent analyst evaluation supports the platform's standing. Held at B for the same reason as its closest peer here: no detection rate, false positive rate or validation result is published, and architectural sophistication is not a measurement.

Third party validation is the strongest of any vendor in this index. The company was named a leader in a major analyst house's anti money laundering evaluation in 2025, appeared in a leading business magazine's fintech fifty the same year, and has won an industry research firm's fraud and anti money laundering impact awards five times.

Two customers are named publicly, a listed business travel and expense platform that deployed detection across its travel, expense and card products in August 2026, and a consumer services marketplace, alongside claimed adoption by Fortune 500 companies and leading financial institutions worldwide. The company has operated since 2013.

No data boundary statement was located, and the question is sharper here than for most because unsupervised learning derives its advantage from finding patterns nobody labelled, which raises directly whether patterns learned in one institution's data inform detection at another. The customer base includes competing banks and payment companies. Nothing states whether models are trained per tenant or across the estate, whether a customer can decline participation, or what happens to behavioural profiles after a contract ends.

No data protection agreement, retention schedule, subprocessor list or deletion commitment was located. The platform holds identity documents from onboarding, full transaction histories, device and behavioural signals and investigation case files for institutions worldwide, which is among the broadest personal data footprints of any vendor in this index, and none of its handling is described in published material.

No attestation, certification, trust centre or enumerated framework was located. Given a customer base including Fortune 500 companies, global banks and a listed technology platform, the underlying assurance programme is necessarily substantial and has been examined repeatedly in private, which makes its absence from published material a disclosure gap rather than a capability one.

Named instruments carry this grade. The platform is positioned explicitly against United States bank secrecy compliance obligations for both banks and credit unions, and it generates suspicious activity report narratives with automated field population and direct filing, which is a specific regulatory artefact with statutory content and timing requirements rather than a generic compliance claim. Customer and business identification workflows are covered as products in their own right. What is absent is any named supervisor and any non domestic regime, despite a customer base described as worldwide.

No individual is scored for credit and the harm is the wrongly blocked customer or the wrongly filed suspicion report, which carries consequences the subject never learns about. Unsupervised detection changes the shape of that risk rather than removing it, since a system that identifies coordinated behaviour without labels can also cluster legitimate customers who share a bank, a device type, a remittance corridor or a migration pattern, and the resulting group flag is harder to interrogate than a rule.

No false positive rate, no analysis across customer populations and no fairness testing was located, and generated suspicion narratives add a further exposure since the reasoning reaching a regulator is machine drafted.

No guarantee, indemnity or falsifiable commitment was located. The institution is well served, with logged agent interactions, auditability and rollback giving it the means to reconstruct and reverse what the system did. The individual has nothing: a customer blocked in real time, declined at onboarding or named in a generated suspicion report has no described route to learn why, contest it or have it corrected, and in the case of a suspicion filing is prohibited from being told at all.

The unsupervised engine is stated to be proprietary and patented, which is a clear ownership disclosure, and the conversational agent layer is described only as built on large language model based agents with no provider, model family or version identified. No hosting arrangement, external data source or subprocessor list appears. For agents drafting regulatory filings and executing approved actions, the identity of the underlying model is what an institution's model risk function would need to assess change and dependency.

The platform is cloud native and built for enterprise scale with real time decisioning, and its unification argument is itself an integration claim, since bringing detection, monitoring, onboarding, case management and decisioning together removes the internal interfaces institutions otherwise maintain between separate vendors. Deployment across a listed platform's travel, expense and card infrastructure in a single unified layer demonstrates breadth across product surfaces. No named core banking, payment or identity system appears and no developer documentation was located.

Cloud native architecture is stated and nothing further: no hosting provider, region selection, residency commitment or private deployment option was located. For a vendor serving banks and credit unions across multiple jurisdictions and holding onboarding identity documents alongside transaction records, residency is a routine procurement question that published material does not answer.

No pricing, packaging or basis of charge was located. The platform spans five previously separate product categories, from onboarding checks through detection to case management, which are conventionally bought and priced separately, and nothing describes whether the unified proposition is licensed as one or modularly, or whether charge falls per transaction, per customer or per institution.

Buyer coverage runs from global banks and credit unions through fintechs and payment companies to digital platforms and marketplaces, with segment specific compliance positioning for depository institutions. Functional coverage is the widest of any financial crime vendor here, spanning onboarding checks for individuals and businesses, transaction monitoring, fraud detection, anti money laundering, case management, investigation and regulatory filing, deliberately unified so risk teams see the whole customer lifecycle rather than working from separate systems. Named use cases range from account takeover and synthetic account creation to buy now pay later application fraud.