Feedzai

Scoring 120 billion events a year in real time cannot be done without models, and the chief science officer argues financial risk is adversarial, with the next transaction far less predictable than the next word in a sentence. Rules are kept, but combined with models rather than standing in for them.

Institutions author custom rules, add third party scores and tune decisioning to their own risk appetite, so the logic is not owned by the vendor alone. Where a transaction is blocked without review, and whether automated decisions are sampled or escalated, is not described.

Importing and exporting machine learning models lets a bank run one its own risk function already validated, and explainability is treated as a differentiator. Accuracy or recall figures, model documentation and a validation package are still unpublished.

Elo, the Brazilian card network, reports a 90 percent fraud reduction across more than 35 migrated issuers, and CoreCard Digital reports a 46 percent cut in declined fraud transactions with a 64 percent detection rate. Payments secured total about $9 trillion, though comparisons against earlier solutions carry no baseline.

Pooling is pitched as the strength: the foundational model is said to match bespoke models on one customer's data and surpass them when trained across several institutions and geographies. Network insights are stated to be anonymized and aggregated, a categorical boundary, though the method, any opt out and retention for contributed data are unpublished.

Transaction records, behavioral biometrics and device intelligence for about a billion consumers feed a shared network layer. The linked privacy policy address leads to a missing page, and the homepage carries no retention schedule or transfer statement for the service.

Page text names no certification, attestation or trust portal, and the security page address leads nowhere. Tier one banks, card networks and acquirers impose heavy assurance demands before transaction data moves, and nothing they would ask for is published.

The European Central Bank selected Feedzai for digital euro fraud prevention, and the scope maps onto money laundering, onboarding and the scam liability regimes reshaping instant payments. As a supplier it needs no license, and no supervisory instrument is named as a design target.

Freedom from bias is a named pillar of a published trustworthy AI framework that also covers transparency, robustness and security, backed by material on fairness in automated decisioning. Results are another matter: the homepage shows no decline rate analysis by population or geography and no fairness testing outcome.

Whitebox Explanations, described in an October 2019 anti money laundering release, show why a risk score was computed, which gives the institution grounds to contest one. A consumer whose legitimate payment is declined has no relationship with the vendor and no described route to see or challenge the outcome, while Feedzai publishes neither an error rate nor a remedy.

An in house foundational model for financial risk is trained on onboarding, payments and anti money laundering workflows across Feedzai's own network. Subprocessors, hosting and the institutions contributing to the shared network are not disclosed.

One RiskOps platform covers card, instant transfer, digital and onboarding flows and enhances existing decision engines rather than replacing them. Mastercard, Fiserv and Jack Henry appear among homepage names, but no core banking or card processing integration is documented.

Global banks and payment networks receive a cloud native service with no stated host, region, residency option or transfer mechanism. For institutions under many data regimes, where data rests is a first order question.

Network scoring through Feedzai IQ, offered to banks large and small through one interface, hints at a lighter entry than the enterprise RiskOps platform. Neither carries a published rate, tier or billing unit.

Coverage runs across the payment ecosystem, from banks of every size and card networks to acquirers, merchants and public authorities. Risk types are spelled out in operational terms, card fraud, account takeover and money mule activity among them, alongside onboarding and anti money laundering.