Online banking electronic payments fraud is on the rise, as fraudsters explore security gaps in the emerging banking channels. For instance, data released by Financial Fraud Action UK, shows that UK online banking fraud losses increased by 38% in 2014 from 2013, and by 64% in 2015 from 2014. The rapid growth of fraud losses calls for an effective e-payments fraud prevention solution and one that can adapt to varying fraud patterns.
In early 2013, we launched the first FICO Retail Banking Payment model on the FICO® Falcon® Fraud Manager 6 platform, a model designed to detect fraud in e-payments (ACH, Swift, BACS, etc). Since then our Retail Banking fraud solution has steadily gained worldwide adoption, with clients in North America, Europe, South Africa, mid-East, and Asia.
Our Retail Banking model is empowered with many advanced analytics patented technologies. As I blogged previously, one distinguishing technology is online self-calibrating models. The technology allows a model to track the distribution of risk features in real time, as transactions stream in, and then uses the updated distribution to identify outliers and normalize the features. This is an important quality in a dynamic environment such as retail banking, area where payment channels keep on evolving.
Another important technology is the behavior-sorted list, which enables real-time learning about customer and account favorite repeated behaviors. The technique enables the model to capture habits of both payer and payee from different dimensions. These could include a list of beneficiary accounts that a payer pays regularly, devices that a payer has used in the past to make payments, foreign countries that a payer paid before and a list of payers from which a payee regularly receive funds.
Based on one study, the transactions that deviate from a customer’s behavior-sorted lists across all dimensions is a minimum of 30 times riskier than the transactions that follow at least one behavior-sorted list. Among fraudulent transactions, about 70% of them have completely new behavior; among non-fraud transactions, about 80% of them follow at least one established behavior.
It’s clear that fraudster attack behaviors are different from that of the legitimate customer. Taking an example of a fraud scenario, where a fraudster gained access to a victim’s account by phishing, the fraudster would most likely initiate payment to beneficiaries that the legitimate payer has not paid in the past, and from a device that the legitimate payer has not used before. As shown above, the behavior-sorted lists feature effectively reduces false positives and enhances fraud detection.
Most recently, we launched our Retail Banking Deposit Fraud models for deposits. The deposit model is equipped with similar patented technologies that are proven to be successful in account-to-account payment fraud solution. Stay tuned for updates on new innovations in our retail banking analytics.
Follow me on Twitter @ScottZoldi.
