Financial crime is a “cat and mouse game”, says Sjoerd Slot. His company, Fraud Dynamics, focuses on detecting financial crime early through providing its largely bank and payments processor client base with the analytical and management tools to get the job done and stay ahead of the game.
What banks can bank on
The constantly-in-flux landscape of financial crime presents a significant problem for banks. For example, there are a myriad of ways to pay, and clever criminals are always one step ahead in finding ways to circumvent detection. This is, Slot remarks, in contrast to an earlier time where recognising fraud used to be clearer.
As the payment landscape changes, so have the risks, necessitating that fraud detection platforms are also flexible. At present, “banks have already invested heavily in many monitoring systems ,” says Slot, so the solutions Fraud Dynamics delivers, instead, integrate into these current systems and are based on anomaly detection models.
Models for fraud detection: less false positives
Typical fraud detection is done based on using established criminal behaviours as guidelines. The problem with this approach is that while companies are modelling behaviour based on the past, criminals are on the move. This lack of agility means less effective fraud detection because systems cannot adjust fast enough to what is occurring.
Flipping the perspective and basing detection on anomalies from the “normal” customer is more effective, which is the type of detection analysis Fraud Dynamics provides. “If you want stable detection,” Slot says, “you need to look at stable factors. Your ‘normal’ customers don’t have the same incentive to constantly change their behaviour like criminals do. Therefore, you have to consider anomaly detection, which also lowers the risk of false positives.”
Anomaly detection by his company, he notes, is more comprehensive than basing detection on individual risk indicators because it takes into account every type of behaviour instead of looking for specific “signatures”, which necessitate constantly updating systems with new rules. Thus, proper anomaly detection gives a more well-rounded picture of what is occurring.
Regarding regulations, however, he asserts that models need to take into account and incorporate specific compliance risks as an integral part of the model because, “anomaly detection without understanding the compliance aspects might also create blind spots”. Therefore, regulations play a key role in Fraud Dynamics’ business, as they are a critical component of transaction monitoring. Slot notes that there are several ex regulators and compliance experts on the Fraud Dynamics team. He adds that, as a consequence of this regulatory environment, it is important for companies to be completely transparent in the way they operate, leaving nothing up to chance during audits. This is also translated in Fraud Dynamics’ approach on how to apply AI and machine learning in their models.
A better way to use AI and machine learning
AI and machine learning are used in their fraud detection analytics, but not in the way you might assume: Fraud Dynamics separates the “model generation” from the fraud detection in live environments. “We use AI and machine learning to come up with the right rulesets,” Slot says. AI and machine learning are used at Fraud Dynamics to help identify what normal, legitimate transaction patterns looks like based on what is happening in a company’s current fraud monitoring system with their own data. Generated models can be implemented in their existing detection systems, which means that they do not have to go down the long and costly road of overhauling these systems.
An additional benefit is that this approach creates transparent models that are generated for the detection systems. These have become fully auditable, contrasting integrated AI which often results in a black box. Choosing the correct application of AI and machine learning for a given problem is the key to utilising these tools most effectively, something at which Fraud Dynamics excels. In fact, Fraud Dynamics is currently beta testing the new version of its analytics tool, preparing it for its scheduled February release.
Managing fraud processes
Next to their innovative analytics offerings, Fraud Dynamics also offers fraud management. This tool gives clients a unified overview for the all of their fraud processes that works seamlessly with back office systems, CRM systems, and existing detection solutions. Customer data, alerts, and product information are all made available, reducing the need to navigate between multiple systems and therefore enhancing the quality and efficiency of the investigation process.
By Elliot Lyons, Research Analyst