Using AI and machine learning to fight money laundering

Data scientists have done a great deal for us in the way of making our financial lives easier and, without a doubt, more secure than ever before. These advancements aren’t always used for good, however. While technology has brought the financial world forward in leaps and bounds, it’s worth remembering that this same technology is used for far more nefarious reasons: cybercrime in its many forms.

The Cost Of Money Laundering

Fighting money laundering is a massive, costly mission that needs to be perpetually in motion. For context, anti-money laundering (AML) measures cost European banks roughly €18 billion each year, with their US counterparts shelling out approximately €22 billion annually.

Banks all over the globe need to have their fingers firmly on the pulse of the fintech ecosystem, and that means paying close attention to their transaction monitoring standards. Failure to do so could leave banks with a hefty fine. The last ten years have seen a whopping 90% of Europe’s banks slapped with fines for failing to take the required AML measures – that’s €23 billion on fines alone.

Changes in the tech landscape

Staying at the forefront of tech is crucial for any organisation hoping to remain relevant in a highly competitive industry. AI and machine learning are developing at breakneck speed, with the biggest cloud providers in the industry making them a clear priority over the last few years.

This tech has the potential to completely revolutionise the front and back-end operations of financial bodies across the globe, boosting risk management efforts, maximising efficiency, and reinforcing the effectiveness of financial crime investigations across the board. Embracing this new tech can also minimise costs by helping financial institutions meet regulations more efficiently, freeing up human resources that can be assigned to other vital areas of the business in the process.

Types of AI and machine learning

There are two types of AI and machine learning, each with its own set of pros and cons – supervised and unsupervised. Supervised learning involves using a model trained using data that has already been categorized to raise a red flag on to any transactions that seem suspicious.  In unsupervised learning, what happens is that raw, uncategorised data is introduced to the system, making it start from scratch. By interacting with that data, the system starts to identify patterns indicative of money laundering activities while also creating new ways to sort and analyse data.

The bottom line

As intelligent as this tech may be, it’s only as good as the data you feed it, so investing in talent should remain a top priority. You can’t expect any model employing AI to work without any sort of human input or testing – not yet, at least. Take transaction monitoring, for example. Each transaction needs to be evaluated against a set of risk-based rules. Even the most advanced monitoring systems to date leave banks with a substantial number of false positives, and that’s when the reviewer comes in to cast a human eye over the results before moving forward. In order for any financial institution to get the best out of its human and tech resources, the two need to work hand in hand. Employees need to feel empowered to work with machines and AI, and tech needs a human element in order to continue moving the industry forward.

ComplyRadar to the rescue

ComplyRadar monitors transactions relating to individuals, accounts, and entities to detect suspicious activity – quickly and effectively. Its machine learning transaction filter provides an additional probabilistic layer on top of the rules engine. This not only drastically reduces false positives but also provides additional data output which is not being captured by rules. For example, the machine learning algorithm can identify that a particular new combination of data is resulting in suspicious transactions. Contact us for more information on how ComplyRadar can automatically identify and react to suspicious behaviour in real-time or retrospectively while minimising unnecessary alerts.