Forecasting fraud: Artificial intelligence tools predict potential offenders

The Internet definitely makes life more dynamic, but it also makes things more complicated-and vulnerable. When IP gets involved, a host of security risks pop up. IP telephony, virtual private networks and on-line billing each pose their own set of threats.

If technical difficulties aren't problematic enough, a company's employees pose significant danger. More often than not, employees aid security attacks by unwittingly revealing information that leads to compromised information systems. These attacks, known as social engineering, rely on the politeness of employees as they share passwords, access codes and remote dial-in information.

Carriers may seem more at risk than ever, but the defense tools are getting stronger and smarter, especially in the data mining and data analysis arena. Although they won't keep the perimeters of the network safe, these systems, integrated with artificial intelligence tools, can proactively limit subscription fraud and churn.

"Credit analysis has been a monthly matter, and fraud detection has been a daily matter," said Ted Crooks, executive director of marketing and product management at HNC Software Inc. "Now there are real-time predictive systems that continually update what customers are doing and forecast where they are going and how [a carrier should] respond."

HNC's ProfitMax is one example of this type of proactive neural network technology. ProfitMax records the details of each customer's interactions. The data is updated with every financial transaction, every call placed and at any other interaction point. Using neural network, pattern-recognition technology, ProfitMax predicts if a customer is unlikely to pay his bill or if he may change to another carrier. It can also predict the best time to take advantage of sales opportunities.

Sprint is using HNC ProfitMax to reduce its credit fraud among residential long-distance customers. It works with customer care, billing and other systems and data sources to create predictive models of customer activities.

HNC took the Sprint customers' calling histories over two years and "tortured the data until it confessed," said Crooks. By analyzing the data and using mathematical techniques to associate behaviors and outcomes, HNC let Sprint see clear patterns in the data and predict which customers would not pay their bills.

"We took the history of their business for two years and tried to figure out the human clockwork behind it," said Crooks.

Sprint is using HNC's neural network-based models to detect two types of risk among customers. First, it detects subscription fraud accounts-those likely to never pay. Second, it monitors the ongoing risk of a customer's account-the likelihood that the customer will cover his outstanding balance.

ProfitMax has two other modules that perform transaction level analysis and individual behavior profiles. These modules identify the churn risk and predict the overall, lifetime revenue of a customer.

"Neural network products are the wave of the future. They use new and evolving technologies to detect patterns that are moving too fast for humans to detect," said Charles Cresson Wood, an information security consultant. "Neural networks that can detect fraud in real time could save millions of dollars."

HNC brings this technology from the financial community, where banks have been using its predictive analysis tools since 1990. It has also added neural network technology to its Advanced Telecommunications Abuse Control Systemfraud detection product, which will be released in March. ATACS 5 will prioritize possible fraud cases and keep a record of how the fraud was handled (see figure).

"In some ways, what we're doing is old-fashioned," said Crooks. "In the past, operators knew their customers and used their own common sense to decide the types of behavior that were reasonable. Even though it's mathematically difficult, we're trying to duplicate that common sense."

BELL ATLANTIC PICKS EXCHANGE Bell Atlantic has selected Exchange Applications' Valex customer optimization software to use in its customer relationship management initiative. In the first phase of a multiphase project, Bell Atlantic will use Valex to market products and services to more than 11 million residential customers in seven Northeastern states.

EVOLVING SYSTEMS, LUCENT PARTNER Evolving Systems Inc. has signed a multiyear agreement with Lucent Technologies to provide customer software support. Under the agreement, Evolving Systems will provide its wireless telecom development and integration expertise for Lucent's wireless packet data projects.