Fraud.net

Models do substantial work, with machine learning tailored per customer rather than one shared scorer, anomaly detection and entity risk intelligence positioned explicitly against event based rules that see transactions without seeing intent. The company's own framing of its architecture nonetheless puts three components in equal standing: collective intelligence from the network, machine learning, and a no code rules engine that customers author and lean on heavily. Apply the removal test and a rules driven platform with network data and orchestration survives, which places this with the decisioning vendors rather than the model native ones.

Control sits with the institution through a no code rules engine it authors and configurable dashboards, so thresholds and policies belong to the customer rather than the vendor, and the stated aim of freeing analysts from reviewing low risk transactions implies human review is retained where risk is higher. Investigation workflows and automation of routine analyst tasks are described.

What is not documented is the boundary: no stated escalation threshold, no approval step before an account is blocked, no sampling or quality control over automated decisions, and no described route by which a rejected transaction is reconsidered.

Partial inspectability comes from the architecture rather than from documentation, since the rules layer is authored and therefore fully known by the institution and the machine learning models are built per customer rather than inherited. The published figures are the weakness.

Fraud losses down 80 percent and false positives down 97 percent are strong claims with no stated sample, baseline or methodology anywhere, and there is no model documentation, no detection or recall rate, no validation summary and no stated support for a customer's own model validation.

Two customers are named with an attributed quote, a European financial solutions group describing growth of its buy now pay later business and reduced fraud attacks, alongside a consumer credit provider, and further case studies are described by shape covering a global payment processor, a top tier bank migrating fraud operations and a major retailer. Headline claims are specific, with fraud losses down as much as 80 percent and false positives cut by 97 percent.

Eleven years of operation counts for something in a category with high churn. What is missing is scale and method: no customer count, no volume figure, and no methodology behind either headline percentage.

One design choice helps and one disclosure gap hurts. Machine learning models are described as tailor made per customer rather than pooled, which scopes learning to the institution and is a better default than a single shared scorer.

Against that, the platform is marketed on access to what it calls the world's largest anti fraud intelligence network, with collective intelligence as a headline benefit, and nothing states what a participating institution contributes, whether contributions are anonymised, whether participation can be declined, or how a bank's transaction patterns are prevented from informing decisions for a competitor on the same network.

The platform ingests, enriches and acts on transaction, identity and entity data across a customer base spanning banking, payments and commerce, and feeds a shared intelligence network on top of that. No published privacy framework, retention schedule, subprocessor list or statement of financial privacy service provider obligations was located, which is a gap for a vendor whose own marketing emphasises breaking down data silos.

No trust centre, enumerated certification list, attestation scope or audit period was located in this pass. The platform sits inside payment flows and holds transaction and identity data for banks, credit unions and processors, all of which run vendor assurance programmes that would have required attestations privately, so the published record almost certainly understates the control environment.

Fraud.net supplies technology and holds no licence, the expected posture, and it demonstrates unusual attentiveness to where regulation is going rather than only where it stands. Published analysis engages directly with the 2026 proposed rulemaking that shifts anti money laundering and counter terrorist financing supervision from checklist compliance toward measurable effectiveness, and argues what that changes about the technology institutions need. Product scope covers customer due diligence, entity screening, transaction monitoring and regulatory reporting. No formal admission process is evidenced.

The headline metric points the same way it does across this category: a 97 percent reduction in false positives is the error that costs the institution money and analyst time, while nothing is published about the error that falls on people, namely legitimate customers declined or accounts blocked. Entity screening carries the name matching asymmetry documented elsewhere in this index, where accuracy varies by naming convention and script. No demographic or segment level decline analysis, fairness testing or accessibility consideration was located.

No accuracy guarantee, remediation commitment or published error rate was located. The consequences of a wrong decision fall outside the customer relationship in the usual pattern: a declined transaction, a blocked account or a rejected onboarding lands on a consumer or business that has no relationship with the vendor, is not told a model produced the outcome, and has no described route to see or contest it. The reported reduction in false positives is offered as a benefit to the institution rather than as a commitment to anyone affected by the remainder.

Two elements of the chain are named, the cloud infrastructure provider and a card processing data source integrated directly for merchant risk monitoring, which is more than several peers disclose. The consequential part is undisclosed. The collective intelligence network is the platform's marketed differentiator and its composition is never described, so a buyer cannot tell whose data informs the signals it receives, and no model providers or subprocessors are identified.

The most concrete integration is with card processing infrastructure, plugging directly into a major processor's data to run real time policy checks, speed merchant boarding and produce portfolio level insight, which puts the platform inside the payment rail rather than beside it. A data orchestration layer ingests and enriches customer and third party data across the platform, and the underlying cloud provider is named. What was not located is breadth of documented connectors, with no core banking or case management partners named individually, no public developer documentation and no partner directory.

Delivery is cloud hosted, with the infrastructure provider named openly and credited with faster model training and deployment, which is slightly more disclosure than most vendors here offer. It stops at the provider. No hosting regions, in country residency options, transfer mechanisms, tenancy separation or subprocessor list were located, and the customer base spans multiple jurisdictions including European clients.

No rates, tiers, billing unit or minimum were located. The platform is sold as a consolidation play, replacing separate fraud, anti money laundering and risk tools with one system, and that argument is a cost argument the buyer cannot complete without a number on either side of it.

The buyer segments are enumerated rather than implied, and financial institutions come first: banks, credit unions and global financial services providers, then payments providers broken out as acquirers, processors and remittance companies, then fintechs and digital platforms, then commerce. Naming credit unions separately matters because most vendors in this lane ignore them.

Fraud types are equally specific, covering account takeover, customer due diligence and anti money laundering, new account fraud, payment fraud, synthetic identity and promotion abuse, so a buyer can see immediately whether their problem is addressed.