Learning on the job

A fraud management system that uses an object database for its memory and a heuristic data engine to "out-think" would-be phone fraud perpetrators will soon help cut down on misuse of prepaid calling cards over BT's network.

The system, dubbed Sheriff, which stands for "statistical heuristic engine to reliably and intelligently fight fraud," is designed to allow the carrier to catch cases of fraud as they happen.

Typically, fraud detection and prevention systems have taken hours or days to accumulate sufficient data to identify probable sources of fraud. However, the Sheriff system allows call traffic to be thoroughly analyzed in near real-time conditions, 24 hours a day, 365 days a year and within minutes of the completion of each call.

The system is designed to analyze call trends such as the time a call is made, where a call comes from and goes to, and to whom it is billed. Sheriff compares these trends with an object database provided by Versant Object Technology. This data is then dynamically applied to artificial intelligence applications that are developed in-house at BT and reside within the Sheriff system. The applications check the data against a series of rules and algorithms that look for suspect patterns of abuse or other anomalies.

When suspicious activity is identified, BT's special fraud case investigators can use Sheriff to make an informed decision about what to do next.

The artificial intelligence component of the system enables Sheriff to vary what it looks for, making it difficult for potential phone defrauders to anticipate what carriers are watching out for.

"Traditional fraud systems are rule-based but sound alarms only when certain parameters are reached," said Jonathan Cooper, director of European Telecom Marketing for Versant. "Smart fraud artists learn what these thresholds are and work underneath them. The system's artificial intelligence can look at trends and 'learn' when situations are occurring that require attention, even when they're below threshold values."

The use of Versant's object-based database provided an important part of the system, according to David Woods, Sheriff program delivery manager at BT Laboratories in Martlesham, England.

The structure of Versant's database allows data to be accessed very quickly, and the database is highly scalable, providing the performance needed in a near real-time application.

"Fraud is a highly dynamic, fast-moving and complex activity," said Woods. "This makes its detection and management ideally suited to a object-based application. Versant enables complex relationships to be defined and modeled in detail and directly into the database-relationships between where, when and how a call is placed, how a call terminates and who pays." Traditional relational databases have difficulty with this, he said.

The object-based database also provides the flexibilty to accommodate changing requirements resulting from the system's artificial intelligence, which alters the structures of analyses on the fly.

The system, which was originally a joint effort between BT and MCI, is completing trials now. It examines between 10 million and 15 million calls each day and is planned as the basis for fraud prevention for all future BT voice services, Woods said.