Applying artificial intelligence to big data for fraud mitigation and prevention
The card payments industry is in an ongoing and eternal battle with fraudsters. As financial institutions take steps to close off one avenue to the swindlers, new techniques to cheat the system are uncovered – and so the game of cat-and-mouse goes on. The fraudsters get smarter and arm themselves with increasingly advanced tools while the banks play endless catch-up.
The Fraud Arms Race
In the digital age, computer processing power cuts both ways: it assists banks in their analysis and detection of fraud, but it also opens up new avenues to cheat and steal. The big difference between the warring parties is that the fraudsters move and evolve in real time, whereas fraud mitigation and prevention is typically response-driven.
Neural networking techniques and the use of “big data” analytics have significantly improved the detection and inhibition of fraud, and helped better uncover past patterns in order to mitigate risk. Modern fraud mitigation systems typically analyze historical information on transaction anomalies over the preceding years to identify bad behavior. From there, rules and strategies are created to ensure the same fraudulent techniques are unsuccessful and bank security is alerted if they occur.
However, past behavior may not always predict how fraudsters will attack in the future. Therefore, as new techniques are developed and shared within the criminal community, financial institutions must continually update their models and strategies to combat new attacks. Unfortunately, there is a time lag built into these systems as the analysis always occurs after an event. Swindlers, meanwhile, take the path of least resistance and increasingly attack in the form of microbursts, where they hit and run before banks have time to know they have been compromised.
The (Artificially) Intelligent Response
While a crystal ball that would predict future fraudulent behavior is unrealistic, we are on the edge of an era of machine learning whereby new payment scams can be spotted and addressed as they happen. Developments in artificial intelligence (AI) have reached a point where systems can see and understand real-time transaction data across multiple networks and make intuitive decisions about what may happen next. An evolution of the big data analysis techniques of today, this will provide security without the time lag – it all happens in real time.
Machine learning adapts over time, and decision-making improves as the amount of captured data grows. By understanding the behavior patterns of all parties involved, machine learning techniques can create models of good and normal conduct, and thereby immediately spot anomalies and highlight them as potentially fraudulent behavior. Of paramount importance is the need to examine transactions from all angles and multiple sources across the entire lifecycle of a transaction: individual cardholders, merchant origination points, across the networks and all the way down to financial institutions’ internal environments. The AI learns which merchants, cardholders and ATMs are problematic and scrutinizes them more rigorously.
Within the cards world, machine learning can be applied to many diverse areas of fraud: card not present, account takedowns, card replication at card present, card skimming devices, card cloning, etc. Machine learning understands global patterns and trends at the network level and finally allows us to do something more with big data than just number crunching in the background.
With a true 360°, end-to-end perspective, machine learning can recognize potentially fraudulent trends and patterns as they happen. With machine learning, there is no more rule-writing based on deep analysis of historic data. Instead, systems learn what good and consistent behavior looks like and then recognize breaks from that. This leads to improved and more nuanced notifications and alerts when transgressions occur. Patterns can be spotted on the fly and a higher score (potentially more risk) is given to potentially bad behavior that may trigger blocks on suspected transactions.
A New Era in Fraud Prevention
Machine learning promises to offer financial institutions a major boost in the arms race against fraudsters. Using algorithms that iteratively learn from multiple sources of fraud-related data, it enables systems to find hidden insights and trends without being explicitly programmed where to look. That way, it adds to established big data analysis and neural networks and allows banks to react in to a fraudulent transaction as it happens.
The beauty of machine learning is that it can understand an evolving criminal trend and stop it immediately. There is no longer a lag between identifying a fraudulent practice and developing a strategy to stop it – the door is slammed shut immediately. And by examining the entire payment from origination to settlement, the wider perspective offers complete, end-to-end security. Plus, machine learning evolves alongside the fraudsters, automatically, and without the need for costly and time-consuming human intervention.
Things Can Only Get Better
Machine learning techniques are still in their infancy within the world of fraud mitigation, but the technology is set to mature fast. We can expect to see more systems rolling out later this year and into 2017. As a word of caution, however, there are no suggestions that machine learning will eradicate fraud altogether. It is simply the next evolutionary step in the eternal fight against those trying to game the system.
Some say it brings fraud prevention into the real-time world, with predictive capabilities that make run-of-the-mill, small-time swindlers a thing of the past.
What do you say?