Fraud Prevention: How AI Helps Track Changes in Customer Behavior

As fraud typologies become more complex, it is harder for firms to ensure they have robust detection practices in place. Yet while some red flags cover many fraud types, precise detection requires a forensic approach. In a constantly evolving risk environment, how can firms ensure they are detecting fraud proactively, efficiently, and accurately?

Common Red Flags

Customer behavior changes are often a core indicator of fraud. For example, in the case of elder financial abuse, the American Bankers Association (ABA) identified 14 red flags to watch out for. These include:

• Transactions suddenly completed for the customer by other individuals – without required documentation (even if they are loved ones or caretakers)
• Account information changes – such as statements sent to addresses not on file for the customer
• Transactions much larger than usual – or that suddenly exceed available funds

Other crimes are more sophisticated, such as account takeover (ATO) fraud. In this situation, a fraudster uses details obtained through hacking or social engineering to gain access to a customer’s account and funds. They then attempt to behave as though they were the customer to avoid detection. Despite ATO fraud’s complexity, certain patterns are commonly visible. For example, changes in a customer’s login behavior could indicate someone else (or even a bot) is attempting to gain access. Other red flags could include changes in typical user routines or IP addresses that don’t match the customer’s normal location.

Similar patterns may occur in the case of digital payment or credit card fraud. In each case, broad changes in historical behavior – like transaction locations, velocities, or amounts – can alert analysts and alerting systems.

Complex Behaviors: Invisible Patterns

Yet many behavioral changes are much subtler, requiring a more granular approach. These changes create atypical patterns that people close to the customer would notice, but anyone else might miss. For example, certain customer habits connect to their psychology, such as times of day for shopping, or saving and investment styles. In particular, criminals committing ATO fraud specialize in mimicking the real customer’s identity. Often, then, the strongest indicator of fraud is a complex combination of signals that, alone, would seem weak.

Conventional rules often struggle to identify such nuanced behavioral changes. And analysts don’t have time to learn the nuances of how every customer behaves.

Using Artificial Intelligence in Behavioral Analytics

How, then, can a fraud prevention team dealing with large volumes ofcustomer profiles detect patterns that might be invisible to those unfamiliar with individual customers’ personalized patterns? How can individual analysts hope to put together complex, weak signals that might fail to trigger traditional rules?

Such hidden and interconnected behavioral anomalies require solutions with enough power to detect patterns at a large scale. Using machine learning, behavioral analytics can connect seemingly unrelated data points in a  customer’s profile – even when faced with multiple accounts and distinct patterns. Armed with powerful tools, fraud and risk teams can detect patterns invisible to the naked eye, helping them stay ahead of complex fraud typologies.