Business Case: Credit Card Fraud Detection
E-commerce: Reducing credit card fraud efficiently via classification
Task
Credit card fraud is a real problem for numerous internet merchants and has caused huge financial damage to online shops in recent years. Various security procedures such as 3D Secure and SecureCode help guard against credit card fraudsters. Integrating these procedures into the order process curbs credit card fraud substantially. But this increase in security creates significant disadvantages for shop owners: The necessary extra step in the order process makes lots of customers uneasy or even scares them away completely. The revenue lost as a result of this can quickly exceed the savings realized from cutting the number of fraud cases. An intelligent algorithm can prevent this problem. Such an algorithm detects “suspicious” orders and deploys the additional security features only for these fishy orders.
Data
Providing the basis for distinguishing between suspicious and unsuspicious orders (so-called classification) is anonymized customer, shopping cart, and tracking data. Online merchants need to provide information on a sufficient number of cases – especially cases of fraud. The main challenge in this phase is to identify and establish the attributes that reliably and selectively point to cases of fraud.
Analytics
Our task consists of building an intelligent classification model capable of optimally distinguishing between cases of fraud and legitimate orders (so-called discriminant analysis). This requires using every trick in the book. The intelligent classification model gives every order a score, which serves as an indicator whether fraud is occurring or not. We then fine-tune the model to determine the optimal threshold value for deploying the security procedure. This calculation takes into account not only a projection of the financial damage from undetected fraud, but also the potential losses arising from an unnecessary use of the security procedure.
Solution
And now it gets interesting: After being implemented in the internet merchant’s shop system, the classification model decides independently and in real time whether an order must undergo an additional security check during the credit card payment process. With the help of our real-time solution, which we tailor to each shop’s individual requirements, our clients have been able to realize six-figure savings annually.