The False Positive False Negative Dialectic

Unlike neural networks, payment system filters do not look at specific actions of consumers and then make decisions based on individual historic patterns. Payment system filters examine the minutiae of specific payment hubs and detect behavior unlikely (the more unlikely the better) to occur in that exact circumstance. Earlier blogs showed focusing on payer location can show impossibility in one case, and an attempt to deceive in another case. Detection of facts like these form the foundation of low false positive results, which I believe, should be the basis for flagging transactions as fraudulent.

There is room for disagreement with regards to low false positives. Some reason that it is better to have more false positive than false negatives since the latter create the greatest protection against financial loss regardless of inconvenience to payment system users. I think fraud detection implementers should have lots of tools in their belt and just like choosing between power and economy settings in some modern cars, detection implementers should be able to set the percentages based on perceived threats, customer requirements, and real cost.

Consider insurance companies making payments after a storm. Concern for fraudulent claims versus concern for rapid payment is different than it is for a claim made with few others in like circumstances. Events like the World Cup now transpiring in Brazil force a surge of value throughout the entire payment system increasing backpressure and the request for speedier authorizations and the subsequent increase to vulnerability to fraud.  Increasing throughput for short periods makes little economic sense yet there it is, increase vulnerability, or increase throughput. Without adding hardware, the only choice the local industry can do to increase throughput is loosen fraud detection parameters. The phenomena works with any surge of human activity, be it security personnel on duty during rush hour at the airport, or firemen on duty during the Fourth of July (US Independence Day celebrated with Fireworks). The choices are few: spend more money, increase the risk, or increase the efficiency of response methods.

Constructing filters for payment systems require domain experts on specific incarnations of payment hubs. For example, a government payment hub payee will have finite ways for disguising false claims. The difference in data between a false claim and a legitimate one may not be discernible at all; however, only people intimately familiar with the complete data set for the entire transaction can create filters to separate the two.   Once in place filters can work in conjunction with other detection methods creating more false positives and less false negatives. Operators of payment systems can then throttle throughput based on management requirements of user convenience and vulnerability. Initially the requirements for a domain expert increase the cost of the entire system and  justifiable only by cost saving from reduction in false payouts.

Next Blog: What makes international Retail Payments so risky and expensive?