Fraud.net

Supervised learning, anomaly detection and graph neural networks do substantial work, with generative AI alongside. Network intelligence and a heavily used no code rules engine stand level with the models, so a rules driven platform with network data would survive their removal.

Thresholds and policies belong to the institution, with configurable dashboards, case management and investigation workflows, and freeing analysts from low risk reviews implies people stay on the higher risk ones. Escalation thresholds, an approval step before an account is blocked and sampling of automated decisions are not documented.

Rules are authored and fully known by the institution, but fraud down 80 percent and false positives down 97 percent carry no sample, baseline or method. Model documentation, recall rates and support for a customer's own validation are unpublished.

Fareportal reports $100 million in new sales and fraud down 30 percent within 90 days, and Arvato and Countingup are named on agility and efficiency. As a travel business, it leaves a bank wanting a reference from its own segment.

Models are tailored per customer, which scopes that learning, but the privacy policy says member data is combined with every other member's to generate results for the Global Anti Fraud Network. What is anonymized, whether a member can decline and how one bank's patterns stay out of a competitor's decisions are not stated.

Personal and transaction data that members submit about their own customers through the API falls under the service privacy policy. Subprocessors, retention periods and the Gramm Leach Bliley Act are not addressed.

SOC 2 Type II, ISO 27001, PCI Data Security Standard and NIST 800 53 alignment are listed on the trust center, with the SOC 2 and PCI documents released on request. Dates and audit periods are not shown.

Customer due diligence, transaction monitoring and regulatory reporting sit in scope, and published analysis engages with the 2026 proposed rulemaking moving supervision toward measurable effectiveness. Supplier status means no license is claimed, and no register entry is evidenced.

False positive reduction, the error that costs the institution, is the headline, while the error that falls on legitimate customers declined or blocked goes unmeasured in what is published. Entity screening carries the usual naming convention and script asymmetry, and no segment level decline analysis or fairness testing appears.

Members' customers can ask what the company holds about them and have it corrected under the privacy policy, last updated February 1, 2025. Someone declined or blocked is not told a model decided and has no route to contest the outcome itself, and the vendor publishes no accuracy commitment.

Infrastructure runs on Amazon Web Services, and a card processing data source is described for merchant monitoring. Composition of the Global Anti Fraud Network, the marketed differentiator, is never described, and model providers and subprocessors are unnamed.

Card processing infrastructure hosts real time policy checks and merchant boarding, putting the platform inside the payment rail, and developer tools and release notes are linked. Neither the card processor nor any core banking or case management system is named.

Amazon Web Services hosts the service, and data sits primarily on servers closest to the customer, with temporary transfers possible during outages. Regions, an in country commitment and tenancy separation are unpublished, and an on premises option is not described.

A free demo is the way in, with no rates, tiers, billing unit or minimum published. The case for replacing separate fraud, money laundering and risk tools with one system cannot be costed without a number.

Financial institutions lead the segment list, with credit unions named separately, followed by acquirers, processors and remittance companies, then fintechs and commerce. Fraud types are listed specifically, from account takeover and synthetic identity to promotion abuse.