Growing Need for Real-Time Fraud Identification
Fraud attacks are getting to be more sophisticated – as technology evolves fraudsters have elevated their game on payment fraud and money laundering. With access to faster and cheaper computing, fraudsters have shifted their targets to more profitable weaker points in the financial services chain. Sixty-five percent of organizations with annual revenues of at least $1 billion were victims of payments fraud in 2014 compared to 56 percent of companies reporting annual revenues of less than $1 billion(1).
Newer business models are constantly evolving – from instant delivery of goods to virtual cash to digital downloads. However, the growth in opportunities has led to a corresponding growth in online fraud and fraud losses particularly in ecommerce where it is 7 times more difficult to prevent fraud than in the person(2) According to LexisNexis Fraud Multiplier, in 2015, every $100 of fraud costs a merchant $223 in true costs.
The ever-faster, ever-bigger cycle of attacks leads to a number of consequences:
- Magnitudes of attacks are exponentially higher Fraudsters are employing distributed networks, internal knowledge, big data, and even machine learning to easily detect vulnerability and maximize the size of the attacks.
- Weakest links create the most exposure Financial systems are interconnected and consist of a long value chain, a networked ecosystem of multiple entities connecting buyers and sellers. Fraud flows to the least-protected components.
- Unexpected attacks can be unsettling and disruptive Organizations can go from not having a fraud problem to being devastated in just a few days (e.g., Target, Neiman Marcus)
Fraud solutions needs to more sophisticated to keep in pace with the fraudsters and react within the short time fraud attacks happen to when they are discovered. Organizations that want to defend themselves against fraud need to have a superior, faster-learning solution that can constantly evolve yet is easy to use and maintain.
Machine Learning Today
Machine learning as a data science to uncover patterns and hidden insights is not entirely a new concept – It has been in play with the use of neural networks starting in the 1980’s. The question therefore is, “Why is there a big buzz around machine learning today?”
The answer lies in the fact that advancement in technology and science has enabled game-changing differences in how machine-learning algorithms have evolved and is being applied.
For example, traditionally, human-generated rule sets were the most prevalent approach in fraud management and still continue to be in practice today. But the quantum leap in computing power and availability of big data over the last 5 years has disrupted how data is being used to identify and prevent fraud. Machine learning uses artificially intelligent computer systems to autonomously learn, predict, act and explain without being explicitly programmed. Simply put, machine learning eliminates the use of preprogrammed rule sets – no matter how complex.
Machine learning enables:
- Real-time decisions: Advances with in-memory, event streaming technology allow risk scoring and decision making in the sub-second range (i.e., ultra-low latency).
- Big Data set processing: Advances in distributed data processing allow analyzing more data while still maintaining realtime decisions without trade-offs between data and latency.
- Reduced cycle time: Learning cycles are continuous unlike batch learning where models become out-of-date; With machine learning, the same transactions being scored also update/teach the machine learning models.
- Increased effectiveness: Extremely subtle patterns and variations can be detected and delivered (e.g. precision, recall) better than humans in many tasks.
- Error-free processing: Enormous amounts of data can now be processed without human-bias or error.
- Cost efficiencies: Address long tail “corner case” distribution.
Application of machine learning has redefined previous strategies and tools in fraud management delivering benefits that were previously not possible with traditional methods.