Featurespace, the world leader in machine learning fraud and financial crime prevention technology, has been awarded two new patents for its innovative technology that will help fight burgeoning levels of global financial crime. The first is for their Fragmentation Engine, and the second for Sandbox Layered State.

Featurespace’s patents are used to combat financial crimes linked to card-based transactions, account-to-account transactions and fraudulent apps that are used falsely to check customers’ details when opening a new bank or customer account.

The need for the patents is timely, as financial crime that impacts business and society is set to rise due to the expected global recession. Trade body UK Finance predicts an ‘epidemic of fraud’ happening this year. Additionally, the United Nations estimates that money laundering costs up to US$2 trillion each year, undermining economic prosperity and financing organised crime.

Dave Excell, Featurespace Founder, commented: “The power and accuracy of our technology in reducing the instances of fraud and financial crime is well proven. What we do is make the world a safer place to transact by preventing fraud from happening in real time. Our patented technology changes lives because fewer people will fall victim of fraud in the future.”

Fragmentation Engine – high event throughput, low latencies

Featurespace’s technology performs real-time risk scoring interventions in the middle of the global payment flow. These are challenging processing conditions for three reasons: decisions must be timely so that they can interdict real-time payments, the technology must keep a vast number of behavioural profiles fresh with the latest transaction information, and the technology must be robust as it’s depended upon within the global payments flow. This translates into a requirement for reliable high throughput and low latency event processing.

Featurespace’s fragmentation engine patent is an innovative solution to these simultaneous event processing requirements. It works by keeping local copies, or caches, of customer behavioral profile data on each processor node and performing a global all-reduce process frequently to keep each processor's view of the world in sync. This innovation benefits the payments industry by enhancing the quality of data used by Featurespace’s Machine Learning models resulting in industry leading, risk detection accuracy at scale.

In practice, Featurespace’s technology compares and crunches vast amounts of data – in 10-30 milliseconds – to find anomalies and changes that flag irregular payments. For example, this could be an automated attack against a financial institution with bots attempting to uncover valid card or account details with an available balance.  

“One of the primary reasons for banks putting their trust in Featurespace is our ultra low latency real time scoring capabilities. Our fragmentation engine allows us to score 10s of thousands of transactions per second in real time with a latency between 5 and 15 milliseconds” said Matt Mills, Featurespace’s Chief Commercial Officer.

Sandbox Layered State

Featurespace’s Sandbox Layered State enables the addition of new behavioral elements to existing behavioral profiles with zero impact on real time processing flow. It’s like being able to add new personal information to an existing contact in the phone book of your mobile phone when the person calls you – with no degradation of the real time performance.

The invention is used to provide two critical change management operations, both of which are a must for mission critical systems such as those found in payment processors and banks: 1) A/B testing of change versions, and 2) backfill and promotion of new versions of behavioural profiles.

Financial Institutions use this innovation to progressively roll out enhancements to their existing fraud prevention analytics, for example by adding device information and profiles to an existing set of profiles based entirely on transaction data. Financial Institutions also use this innovation to run two versions of behavioural analytics in parallel. They could run a live A/B test of the analytics and then seamlessly promote the more effective version to production.

Critically, Featurespace’s invention achieves these operations with hardware efficiency, making them suitable for production systems running on minimal hardware on premise. To achieve this, the invention uses innovative memory management and search processes involving a hierarchy of copy-on-write snapshot caches and time-aware overlays. This enables the changes to be made ‘just-in-time’ without causing ‘stop the world’ pauses to update the behavioural profile databases.

What sets Featurespace’s patents apart from competitors is the company’s ability to process real-time analytics in the machine learning space without degrading the analytical quality needed to detect fraud.

Featurespace’s patented technology is already in use in the marketplace with some of the biggest banks and payment processors in the world where these innovations are having real world impact, for example: Worldpay, TSYS Global Payments, NatWest, and eftpos. Other solutions might be able to match Featurespace’s speed, but they achieve this with less accuracy – leading to fraud slipping through the net. This can be best compared to the tyres on a high-performance racing car; competitors’ tyres can handle speed, but struggle for grip when turning a corner, leading to a crash. Featurespace’s tech has both the speed and the accuracy needed for successful fraud prevention.

Featurespace’s latest patents are intended for use by regulated financial services, payment service providers (PSPs), banks and businesses, in the international fight against financial crime and fraud. Featurespace’s patent technology is designed with legal and regulatory frameworks in mind, such as the EU’s Artificial Intelligence Act and the UK Bank of England’s Artificial Intelligence and Machine Learning regulation.