Fraud Analytics: We Are Just Scratching the Surface
I have alluded to fraud analytics a number of times in the book without explaining in any detail what exactly I mean. In this chapter, we will examine the origin and evolution of predictive (or detection) analytics in fraud management.
A few decades ago, when banks first started accumulating data, databases quickly followed. Reporting became popular very quickly as well. In the absence of any information, when data is accumulated and some reports are run, it feels like lighting a candle in a dark room. The information seems extremely useful, and numbers-savvy people start looking at the reports and how some of the quantities (variables) in the report vary proportional to others. There is sometimes a direct correlation between fields; sometimes there is implied correlation. When domain experts start to see these correlations, relating these numbers to their own experience in risk management, they venture to write some rules to manage, say, fraud risk. These rules work effectively for a while but since they are rules, fraudsters become very good at figuring out what sort of fraud the rules are designed to stop. Fraudsters then figure out a way to fly below the radar. If the rule cutoff is $200 for a cash withdrawal, for example, the fraudster experiments with various amounts and determines that withdrawing $190 is a safe option. The domain experts see this and start designing rules that would catch this. In effect, it ...