By Eliano Marques, Global Director of Data Science, Think Big Analytics, A Teradata Company

Fighting fraud is a very sophisticated challenge, with investigations into money laundering increasing significantly year upon year, and many banks paying large fines for the failure to deploy sufficient anti-fraud measures. Earlier this year, one of the largest tier one global banks was fined more than £500m for what has commonly become known as ‘anti-money laundering failings’ for transactions that were highly suggestive of fraudulent activity.

Fraudsters are becoming more and more advanced, but luckily, artificial intelligence (AI) and deep learning are giving banks’ fraud investigation teams the ability to investigate crime in real-time, tackling complex, highly varying functions.

Having a system that is capable of learning very quickly and understanding new patterns that are not in line with genuine transactions is a crucial aspect of fighting fraud. Trying to recognise these patterns may be fairly complex and highly nonlinear, but this is a field where deep learning and AI shine by drawing quick conclusions against these multifaceted connections.

Google’s Alpha Go is a good generic example of bringing the initiative to the next level – implementing a deep learning type of model that learnt from its opponent and adjusted behaviour while playing against new opponents to win. It’s this same model is now being applied to investigate fraud in banking, with these new algorithms beginning to win the battle with fraudsters.

Deep Learning is Key

In the banking world, there are a few key areas where AI and deep learning can help. Deep learning can recover complex patterns in data, which can be particularly useful in uncovering fraud.

Deep learning sets itself apart from conventional learning techniques with its very specific techniques and ability to discover complex patterns and has proven a great tool for detecting these types of intricate connections.

The focus must be on finding anomalies in the data. For example, for every transaction that is executed, it’s vital to assess historical transactions to detect which customer transactions do not fit into a typical history from a purely behavioural perspective.

However, we are not able to learn from purely historical data quickly enough because as soon as we are able to figure out which transactions are fraudulent, fraudsters adjust immediately and try a different approach.

As a technical issue, fraud is essentially like a game theory type problem in that the strictly dominant strategy will always adapt to survive and find new ways of attempting to succeed in the same activity. This is where real-time engines are key, and where banks can implement game-changing solutions in the fight against fraud.

AI Driven Real-Time Fraud Solutions: Banking’s Secret Weapon

Many businesses are struggling with legacy technology that simply doesn’t deliver the insight needed to fight suspicious activity, making them unable to move quickly enough to analyse transactions and to identify money laundering and other fraudulent activity.

However, some banks are developing real-time solutions in the form of state of the art, AI driven detection engines. These new engines use machine learning to hook advanced analytics blueprints up to incoming transactions.

Having a re-enforcement type of algorithm is of vital importance – it’s an algorithm that is capable of learning in a fast paced environment and adjusting to fraudsters’ techniques in real-time.

When in place, AI can give responses back to score transactions using deep learning algorithms, delivering immediately available and actionable insight in real-time. So, when a customer is trying to make a purchase using a debit or credit card, the detection engine can score transactions within 0.3 seconds, flagging fraud or approving genuine transactions without interruption to purchases.

While it’s great for customers, it’s also very beneficial for banks. It’s important to remember that while extremely impactful, fraud cases are still relatively rare. When fraud detection methods aren’t sophisticated enough to keep up, they may start to flag up to 99% of all transactions as potentially fraudulent.

Often, it’s the case that only 0.5 per cent of transactions are fraudulent, leaving fraud investigation teams tasked with the highly repetitive and time-consuming work of investigating what end up being ‘false positives’. When better detection is in place, false positives are significantly reduced, leaving teams free to be deployed on higher value, more meaningful work.

Beyond reducing false positives, these systems are catching more of the fraud that is actually occurring, and also meeting the prevention measures set by regulators. They are also helping banks avoid hefty fines and business lost through reputational damage to the brand.

While all banks are facing the same challenges surrounding fraud, it’s clear that some are making better use of the new analytic technologies available to increase the efficiency with how they investigate fraud and comply with standards. Amid a landscape where banks often feel hampered by the rising costs of regulation, it’s safe to say that the tactical use of AI and machine learning will likely soon become a more common fixture in the war against fraud.