Catching the frauds and how!

Catching the frauds and how!

SHARE

ESA and Feedzai

Credit card processor improves fraud detection by 27% by deploying a fraud risk prevention solution using space and location-based technology

Each year, $11.4 billion is lost to credit card fraud. As cybercriminals grow more sophisticated, that number is likely to grow. Portuguese scientists have created a 21st-century way of detecting banking fraud with the help from the European Space Agency and today, every electronic purchase in Portugal runs through their software. Globally, Feedzai products screen about $229 billion-worth of payments every year.

But what do space missions and software designed to find thieves have in common? More than you might think — in addition to high-tech hardware, space missions require a great deal of sophisticated software.

Fraud detection and space mission software face similar challenges. For one thing, both need to process huge amounts of information in real time. “If we talk about a bank, you need to process thousands of transactions every single second,” says Paulo Marques, who was an ESA consultant before founding Feedzai in 2009.

At ESA, Paulo and Feedzai’s Nuno Sebastiao called on high-performance computing techniques to create virtual satellites: “Clusters of computers pretend to be everything involved. A computer acts like a spacecraft.”

In bank fraud detection, as in space, the software must recognise anything that is out of the ordinary. In space, an unexpected change in temperature could indicate a crack in the wall. In banking, anomalies often point to fraud: if a petrol station suddenly starts generating sales figures like those of a luxury car dealership, it is a sign of trouble.

However, there are differences. While hard-and-fast rules are set to detect an anomaly in space, fraud requires decisions on a case-by-case basis. A sudden temperature change in a spacecraft is always a problem, but each bank customer has his own, individual habits.

As a result, the software must recognise what is normal for a business-owner and what is normal for a teacher, based on the past practices of each, before it can identify any odd behaviour. To make this possible, Feedzai came up with an artificial intelligence software system. “

We developed software that can process a huge number of transactions,” said Paulo. This software can look at every transaction a customer has made for the last four years.

By applying both ‘machine learning’ and ‘big-data techniques’ to look at all the data, the software learns to distinguish fraudulent-looking from non-fraudulent-looking transactions. “The software creates the rules,” adds Paulo.

As world continues to adopt electronic payments, our lives increasingly migrate to the Internet. As a result, customers (and criminals) leave behind a growing trail of digital exhaust. Using geo-location data from mobile devices or IP addresses, Feedzai can determine each consumer’s spending pattern and predict normal behaviour. While techniques like velocity rules that rationalise time and physical distances between purchases are not new, Feedzai adds modern big data computing and data science techniques, which allow for machines to learn from the data. Consequently, clients like banks, retailers and payment networks are able to keenly focus on their customers in order to provide a happy commerce journey, whether that is online or in a store.

A branch of BPI

Feedzai’s software is robust. Tracking over 300 variables per person, it creates a very detailed, individualised spending profiles for as many as 20 million credit cardholders per system. “In total, we are tracking over 5 billion variables continuously.”

“It is like having 500 very intelligent people looking at every single transaction and making a call based on their experience if that is fraud or not. It is a huge amount of computing power.”

Cerqueira from Instituto Pedro Nunes, the Portuguese broker in ESA’s Technology Transfer Network part of ESA’s Technology Transfer Programme, believes Feedzai’s technology will mean savings for banks, as well as improved customer loyalty: “Feedzai’s machine learning models and Big Data science are able to detect fraud up to 30% earlier than traditional methods, and illustrate how the competencies developed at ESA research centres can be useful to other sectors.”

Improving fraud detection by 27%
A European processor serving the commercial banking industry (name withheld owing to client confidentiality) sought to quickly deploy a solution that would stop a greater percentage of fraudulent transactions, while reducing false positives whose manual reviews are costly and frequently time consuming. With 5 million transactions per day to validate, the company’s general manager understood that a new solution’s time to market would significantly impact the business.

With Feedzai, the company achieved the fraud prevention results they sought, and more. The fraud protection was improved by 27% while limiting false positives to 20%. “Card not present” transactions, always a challenging use case for payment processors, were validated against a year’s worth of customer, card and merchant behavioural data in 20 milliseconds. The company also found that with the software’s extra large datasets and powerful engine, they were able to halt instances of ‘first fraud’, adding value to its financial institution customers.

Feedzai Fraud Risk Prevention was deployed on premise at the processor within two weeks using existing offthe- shelf hardware, assuring that processing and energy costs were kept to a minimum. Capable of processing up to three years of historical data for cards, cardholders, and merchants, Feedzai also enabled the processor to stream real-time transactional data, allowing each transaction to be evaluated from multiple dimensions. Machine learning classifications were then applied to the system, eliminating the need for the costly and time consuming manual updating process of other rules-based solutions. With real-time transactional data continually updating each record, the processor is now able to automatically retrain its fraud algorithm based on up-to-the-second information, positively impacting the detection of fraudulent “card not present” transactions, as well as card-cloning and merchant fraud schemes.

Feedzai Fraud Prevention now allows the processor to bring together past data, present anomalies, and future predictions to uncover, prevent and manage 80% of fraudulent transactions for a tenth of the cost of competitors’ solutions. Its false positives rates are now consistently at 20%, trimming the costly review process for its client companies while enhancing the experience of its cardholders. “Feedzai prevents $9.5 million per year in credit card fraud losses for us by uncovering and cancelling fraud in real time,” says the processor’s general manager. The company now analyses years of historical data, applies machine learning classifications, and determines whether a transaction is fraudulent in under two milliseconds. Because Feedzai classifies and delivers fraud scores so quickly, it means a fraudulent transaction can be blocked even before the bank authorises it. By detecting and preventing these ‘first time fraud’ transactions, the processor realizes a benefit previously unheard of in the industry. The bottom line for processor was that Feedzai delivered a two-month return on investment.