New Fraud Prevention Model

Digital payments and alternative payment types have led to a massive increase in the amount of information credit unions need to evaluate when discerning the nature of a transaction and its legitimacy. With so many variables to analyze, it is becoming increasingly difficult to set up the right controls to minimize risks and prevent, or substantially reduce, fraudulent activity. Fraud remains a daunting issue, the price of which is not simply financial loss, but loss of customer confidence and damage to brand integrity.

Most modern fraud prevention and detection systems are rules-based engines that can be set up to be as simplistic or sophisticated as each individual CU requires. Rules can be built around the analysis of transactions. For example, in the simplest case, a rule could consider the improbability of a transaction being legitimately made in France, when a purchase was made an hour earlier in Boston where the cardholder lives.

Object behavior is a more sophisticated method of analysis in which an object could be a card, account, retailer, terminal, country or merchant. Statistically analyzing the object’s activity over a certain period enables the tracking of changes in the object’s behavior—for example, criminal activity at a specific ATM.

Machine learning is the latest method of fraud prevention and detection. It incorporates artificial intelligence where, in essence, the fraud system teaches itself. This is accomplished by using historical transaction data to build a model that can detect fraudulent patterns, using not just transaction flow and object analysis, but also clusters of these objects to identify patterns that wouldn’t be found otherwise.

Cluster analysis divides objects into groups that share common characteristics and determines the relationship between these groups, applying complex logic that goes beyond the realm of human analysis. The subsequent model is then implemented to analyze transactions with much more intuitive behavior-based results.

These machine learning models need to be updated periodically to ensure new fraudulent patterns of behavior are picked up. This means there are far fewer false alarms, and the frequency of actual fraudulent occurrences is substantially lowered.

The success of machine learning methods is fundamentally linked to the volume of data available, as a higher volume of transactions makes it easier to spot less obvious fraudulent trends. However, the advantage for institutions with a relatively low volume of data available to them lies in the detection of patterns. To remain competitive against large banks, CUs are beginning to offer mobile and other transaction channels, opening the door to new types of payment fraud. Machine learning can essentially be used to fill the gaps in current fraud prevention strategies by acting like an electronic watchdog. This methodology isn’t replacing or eliminating the systems already in place, but extending their functionality by adding an additional layer that looks for cluster patterns.

The associated cost of implementing a solution will be based on such factors as transaction volume and vendor choice. The justification is the assurance against the financial losses associated with fraud, and this idea that keeps coming up at industry conferences: Dependant on CU size, using machine learning tools should pay for itself within 12 months, if not sooner, for eliminating fraud. Machine learning methods may not currently be the norm, or be widely used in the industry today, but they are worth exploring by CUs.

The philosophy behind this preventative software is to future-proof fraud systems; fraud is going to shift, with criminals diligent in their attempts to hack systems and find new ways to acquire member data. Machine learning is a methodology that can measure risk and patterns to the extent that it can uncover new types of fraud, from as early on as account opening. This gives CUs the advance notice they need to take a more proactive approach, enabling them to make smarter decisions and create new rules to safeguard their systems.

At a time when financial crimes are becoming more sophisticated, CUs need to be aware of the options available to them to help protect their member data and to both prevent and detect fraud more effectively. Arguably the rollout of EMV will help with card present fraud, however, with the growth in popularity of card-not-present transactions, early identification of potential criminal activity could really make a difference in the long run. Combating fraud isn’t as simple as using one standalone solution. CUs need to look at the bigger picture to decide how they can best protect their members in the long term.

Bethan Cowper is head of international marketing for Compass Plus, a provider of transaction processing, card services, Internet banking, fraud/security, and workflow solutions based in Nottingham, United Kingdom.