The digitization of financial transactions and growth of mobile banking have tremendously increased the amount of information banks need to evaluate for fraud. This presents new challenges to Chief Risk Officers who use human analysis of these large data sets to identify new risk threats. The answer lies in using automated analysis of data, or machine learning, to keep pace with fraudsters and maintain an affordable process.
Machine learning, a branch of artificial intelligence, simply means that the computer can learn from data rather than being explicitly told (programmed) what to do. For example, a machine learning system could learn to distinguish between suspicious account behavior and normal customer actions.Similarly, machine learning can ascertain patterns that signal fraud in account sign-ups to prevent fraudsters from entering the system.
The Evolution of Risk Management
Most lenders have built home-grown systems that use business rules to manage their fraud detection processes. These hand-crafted rules are framed as “if-then” statements. An example would be: “if several transactions are made within a short amount of time in a different state, then send the account to manual review.” These rules have been built and refined with decades of experience analyzing fraud data. Many of the rules are set up to provide additional analysis for unusual account behavior or suspect information at the point of account sign-up.
As the amount of data being produced and the complexity of analyzing has grown to unprecedented levels, the manual process of building and maintaining business rules is becoming more expensive, more time intensive, and less predictive.
Less predictive because criminal elements have become more sophisticated. ThreatMetrix recently reported that over 110 million Americans have been “outed” in the past 12 months (meaning their personal information has been stolen in a cyber breach). Traditional models that rely on normal behaviors can be more easily spoofed.
In response, lenders are beefing up their abilities to verify identity, detect subtly unusual patterns that might indicate fraud, and interact proactively with consumers when fraud is suspected. While there are dozens of third-party fraud prevention platforms to help banks bear the load and protect their customers, most of these systems still rely on human crafted rules that need long periods of analysis to adapt. They are unprepared for the future.
The Future of Fraud Management
So what does machine learning mean for fraud prevention systems in financial services? Machine learning promises to analyze data more efficiently, build statistical models quickly, and most importantly, to react to new criminal behaviors faster.
Faced with the rapid and ever-evolving tactics of today’s fraudsters, banks need every advantage they can get. This next generation of automated model development with machine learning will provide that advantage. Algorithms can learn faster and adjust more accurately than human beings.
The two key components of successful machine learning are:
- The quality of the algorithm used to identify new types of fraud, and
- The quality of the data used to train the algorithm.
Like any software implementation, if any key component is weak, the entire platform will fall short of its’ objectives. Even as banks are recruiting statisticians to manually build the old style models, they will need to find data scientists capable of implementing this next generation of artificial intelligence.
The advent of machine learning for fraud prevention will change how banks manage their fraud risk programs. Human oversight and intuition will remain critical to success, but the machines will increasingly handle the heavy lifting. The growing sophistication of fraudsters continues to push all institutions to adopt more advanced methods. Machine learning is the next step in that progression.