There are several emerging technologies for CNP Fraud Prevention. These new techniques are unlike the traditional, outdated, rules-based, scoring systems, where the fraud manager is required to create risk rules manually. Some companies prefer “Machine Learning” techniques. Others opt for “Artificial Intelligence (AI)” techniques.  There are, however, critical differences of these various systems. For enterprise companies, the seemingly subtle differences can have substantial ramifications.

Machine Learning is computer science that studies pattern recognition, and gives computers the ability to “learn” without being explicitly programmed. This is done by using algorithms that learn from past transaction data to make predictions on future events. A good example of this application is the way Amazon makes suggestions in their “Recommendations for You” section on their home page.

Artificial intelligence (AI) is computer science that performs tasks that normally require human intelligence, such as perception and decision-making. In short, it simulates human reasoning and judgement.

While Machine Learning applications can do a great job mitigating simple fraud, it does not work as well with newer, more complicated fraud tactics. For example, South Africa was once considered a fairly safe country to accept credit card payments from. The Machine Learning system self-created rules which white-listed South Africa and gave the region a positive reputation, based on its past history, even as fraud escalated. By the time the system had caught on, an enormous amount of fraud had been accomplished, and millions of dollars lost. Assuming that the patterns of the past will be repeated in the future does not always work – certainly not with the savvy fraudster. An experienced fraudster works very hard on being innovative and employs new tactics HOURLY to remain undetected. Fraudsters actively avoid being predictable.

Artificial Intelligence, however, has greater success. While it can automate the simple tasks for simple transactions which do not require much human intellect, it can also use computer automation to simulate decision-making on complex transactions without requiring past transaction data or patterns. For example, if an online merchant receives 1,000 transactions per day, 90% (900 transactions) can be automated with a traditional Risk Scoring System or a Machine Learning system. The remaining 10% may, however, require human intelligence to make the final decision on those 10% difficult transactions. They are difficult for Scoring Systems because a manual created rule has not been created. They are difficult for a Machine Learning system because it cannot predict something without previous pattern recognition.

The best way for a merchant to fight fraud is to combine a system or Verification Department framework which automates simple transactions complimented with automation of human intelligence. The automation technology and artificial intellect must also be capable of simulating manual tasks even for transactions that don’t have a previous pattern recognition. Such a synergy is essential to effectively defeat the savvy credit card fraud fraudster.