Fraud detection support system

Customer profile

Play is the most frequently chosen mobile network in Poland. More than 15 million of active SIM cards means that the carrier’s market share is above 28%.

Business need

Internal frauds is a problem faced by all operators, and it significantly affects their annual revenue. In addition to financial losses, frauds can also impact on how a company is perceived by customers and, consequently, lead to reduced ability to increase the service market share. Telecom frauds may take different forms, evolve in time and occur as new phenomena that follow the development and changes in organizations. Quite often, by the time the operator becomes aware of a fraud, it is too late to reduce financial losses and implement relevant remedies. The reporting of basic operational indicators calculated with different technologies and coming from several unintegrated systems is not enough to monitor the occurrence of fraud events and may lead to wrong conclusions.

The scale and cost of fraud schemes force a shift in approach from reactive (analysis of detected fraud cases) to proactive (analysis of cases where circumstances indicate a risk of potential fraud). As the analysis of suspicious events is a time-consuming process, and the distribution and allocation of tasks to operational teams have certain limitations, new methods must be found to support analysts in their daily work and correct risk assessment.

Solution

PLAY, a telecom operator in Poland, has decided to use SAS analytics solutions to support its fraud detection processes in sales network and financial settlements. The following features of SAS software were decisive for selection of this platform:

• Consistency of the technology platform that includes data storage, processing and management layers; fraud scenario engine; availability of advanced statistical analysis and machine learning algorithms; operational and reporting interfaces
• Platform flexibility enabling extension of the solution to cover new areas
• Seeing the big picture of company’s internal processes to grasp the context of observed activities and events
• Hybrid analytics that combines rule-based approach with the use of self-learning models and social network analysis
• Solutions that enable automatic optimizing of analytical model quality
• Substantive and technical support provided by SAS both during and after the implementation process.

SAS® Fraud Framework provides a flexible and comprehensive response to the demand for intelligent tools for detecting events that are broadly defined as fraud. It comprises of components that support users on each stage of the fraud detection process: automation of data calculation (Data Integration Studio); advanced analytics (statistical analysis engine, data mining and machine learning, SAS Social Network Analysis algorithms); operational interface (Investigator’s Panel); and a reporting interface based on SAS Visual Analytics. In addition, the last phase of the project included building of a machine learning model based on a SAS® Viya product, SAS Visual Data Mining and Machine Learning. The solution enables advanced ad-hoc data mining as well as building machine learning models based on the newest analytical methods.

The system in its complete version was implemented in three phases. In the first phase, functionalities related to fraud detection and fraud scenario implementation were provided. In the second phase, the system was enhanced with comprehensive reporting features. In the third phase, new fraud areas were added and the first machine learning model was built to support investigators using the solution.

Business objectives

By implementing SAS® Fraud Framework, PLAY has built a comprehensive fraud detection platform that enables identification and prevention of unwanted events as well as reduction of related losses in selected areas of the company’s business. The key objectives of the implementation are:

• Identify the key fraud schemes to be monitored
• Automate the detection and monitoring of identified fraud schemes
• Support the validation processes by providing the analysts with relevant additional information
• Ensure that monitoring can be extended to cover new schemes
• Provide predictive indicators based on machine learning models to evaluate fraud risks in the nearest future
• Report the scale and progress in fraud detection processes

The solution is used by a unit responsible for the reduction of fraud risks as part of Fraud Risk Management processes.

Benefits

By implementing the solution, the operator was able to reduce fraud-related losses and, above all, shorten significantly the time between the occurrence of a high risk event and delivering the information to the unit responsible for monitoring. The processes monitored by the system include activation, assignment, retention, cross-selling/up-selling and migration as well as financial settlements. The system is also an important operational risk reduction tool. The implementation of the machine learning model will also help to achieve additional benefits such as improved identification of suspicious events. As early as during the model building phase, it turned out that it would outperform the conventional predictive modelling processes by about 30%.

The system will be enhanced to include new areas of data and new fraud scenarios. The development towards the further use of machine learning methods for fraud detection is also planned.