In recent years, growing losses from fraud have cost financialinstitutions millions upon millions of dollars. According to VisaInternational, gross fraud from debit and credit cards is up 29% in2005 over 2004. TowerGroup predicts that fraud from debit cardsalone will surpass $1 billion USD in 2006. Fraudsters have becomemore sophisticated, more organized and more international in scopethan ever before. Financial institutions are under attack on adaily basis and continue to experience losses even with variousfraud control mechanisms in place.

In response to this growth in transaction-based card fraud, mostissuers have made use of neural networks and behavioral models todistinguish normal cardholder activity from fraudulent. Whileneural network technology and scoring models provide the base fornearly all fraud service programs, there are additional areas uponwhich issuers can focus to further improve their profitability byreducing fraud losses.

As effective fraud programs analyze data coming from multiplesources, be it the cardholder, the transaction, the merchant and/orcard networks, it becomes increasingly important to create anddeploy rules as a result of fraud pattern analysis into a real-timeneural network. By using a real-time neural network engine, issuersare able to decline authorization requests for fraudulenttransactions before approvals are provided back to the merchant.Near real-time systems receive transaction data after thetransaction is authorized, which means that at the very best, onlythe second transaction can be declined. Since the vast majority ofcard fraud takes place within the first several hours of the fraudevent, it is imperative that the processor analyzes data throughoutthe day, searching for fraud patterns or information that leads toadding to or modifying the rule set in place.