Machine learning, or the ability for a computer to analyze new information and adapt its calculations accordingly without the need for human input, is one of the most exciting new technologies currently being rolled out by organizations around the world, and financial services is no exception.
One of the most promising applications for this technology in the banking sector is in fraud detection. Being able to spot and act on suspicious activity and transactions in real-time, without the need for human interaction, has become essential for any bank in an environment where card and payment fraud levels are on the rise.
In the US alone, losses from card-not-present fraud reached $16 billion in 2016, according to research from Javelin, with around 15.4 million US consumers affected by this form of payments fraud. This was nearly two million more than in 2015, and the expectations are that these numbers will continue to rise in the coming years, partly due to the widespread rollout of EMV cards in the US. This will make card-present fraud harder to commit, therefore driving fraudsters to the weakest channel – and digital channels provide greater odds of success.
What machine learning offers
In this challenging environment, machine learning fraud prevention technology plays a key role in fighting fraud. On a daily basis, fraudsters make a living on payments fraud and will churn out many attempts before they are successful. To combat this, banks need machine learning in order to successfully block fraudulent transactions during the authorization process.
At the same time, the ever-evolving fraud tactics make it extremely difficult for human operators to spot fraudulent patterns on their own – and this is where machine learning technology offers a significant advantage over less-advanced solutions.
In the past, fraud detection algorithms would be based on specific hard-coded sets of rules and logic questions that would evaluate each transaction to determine its likelihood of being fraudulent. So, for example, purchases made from a different country to where a card is registered may carry a higher risk score, as well as transactions made at merchants that a customer does not normally shop at.
The issue with these rigid algorithms is they do not adapt to new information or emerging patterns unless instructed to do so by a human operator or if the model is refreshed. This can lead to legitimate transactions being declined or new fraud tactics slipping through – and so this is where the power of machine learning comes in.
Beyond basic algorithms
This technology is able to study data points across millions of transactions and adapt its algorithms to take into account new trends without the need for human intervention. So for instance, on a wide scale, if fraudsters adopt a new pattern of behavior that is detected and confirmed across many transaction attempts, a machine learning system can spot this and add it to the list of criteria it checks against automatically.
But machine learning can also work on a much more individual level than is otherwise possible, building up a detailed picture of each customer’s unique spending habits. This means what may be considered unusual and potentially fraudulent behavior for the majority of customers can be dismissed by a machine learning solution that recognizes it as routine for an individual, resulting in a happy customer!
So for instance, if a person works a night shift, they may frequently make transactions in the early hours. Normally, this may be considered as being suspicious and a higher risk score may be calculated, but if a machine learning solution can spot this pattern, it can rate this as a lower risk as it relates back to the behavior profile for that person.
The impact of implementing machine learning technology can be significant. According to one analysis conducted in 2016, implementing machine learning into fraud detection activities could save banks and other card issuers around the world as much as $12 billion a year, cutting the number of undetected fraud incidents by 25 percent and reducing false positives by over 70 percent.