It’s hard for most people to grasp the scale of global money laundering.

The United Nations’ best guess is that between $800 billion and $2 trillion get laundered every single year.

That money flows through mazes and layers of transactions all so criminals can attempt to hide the illegal origins of their funds.

$2 trillion is roughly equal to the annual GDP for Italy or Brazil. Put another way, the entire Brazilian labor force produces as much economic value as financial criminals – arms dealers, fraud rings, human traffickers, supporters of terrorist groups – attempt to hide each year.

That’s the scale and complexity at which anti-money laundering (AML) professionals must work. To do so, they need technology that can sift through mountains of data.

Machine learning has emerged as the technology of choice for AML teams to combat financial crime at scale.

How is machine learning used in anti-money laundering?

Machine learning for AML helps organizations recognize suspicious behaviors and prioritize their responses based on the level of perceived risk. As the machine learning model detects, flags, and prioritizes cases, AML analysts step in to review those alerts manually to make decisions regarding investigations, file Suspicious Activity Reports (SARs) to remain compliant, and assist law enforcement teams to prosecute economic criminals.

To understand how this all works, it’s important to first recognize the role of models in machine learning performance.

Rules versus models in AML platforms

In legacy AML systems, alerts get generated when the system detects activity that breaches pre-programmed rules. Those systems might flag, for example, any transaction above $10,000 or some other regulatory threshold.

There are two problems with rules-only AML systems, however:

  • Comparing real-time activity with prescribed rules generates a lot of false positives. With such a system, an AML analyst might have to field dozens of alerts every day for completely legitimate transactions. That’s inefficient and means analysts cannot get to the real cases, as well as making genuine cases difficult to spot as so many are presented for review.
  • Money launderers themselves change the rules constantly. Having to periodically adjust the AML system to new criminal developments means analysts are always at a disadvantage. They perpetually remain one step behind the criminals.

These problems mean that AML teams relying on rules-only may fail to catch incidents of money laundering or fail to file SARs within prescribed time frames. In both scenarios, the financial institution is at risk of non-compliance and fines.

Enter machine learning models. Machine learning models take huge sets of data, learn from the behaviors encoded in that data, then reveal patterns of activity that indicate evidence of money laundering.

The right machine learning model can supercharge an AML system by:

  • Reducing the number of false positives.
  • More accurately identifying money laundering.
  • Learning on its own rather than waiting for an engineer to program new rules.
  • Ultimately making operations much more efficient and effective.

Model governance in AML

The scale at which AML teams must operate is huge.

Analyzing data and making predictions at that scale is incredibly complex, far more so than humans can achieve without the right technology. But AML technology must still be unbiased. It must operate within the oversight of regulatory frameworks. That’s where the concept of model governance comes in.

Model governance ensures that a machine learning model is transparent and explainable. It ensures the AML teams can explain how and why systems flagged a specific transaction or customer’s activity.

How does AI help AML?

Often, solutions providers will talk about how their systems are built with artificial intelligence (AI) and machine learning. This can be a little confusing.

Machine learning is a subset of AI. Artificial intelligence is a broad term that describes the ways computing technology can mimic human intelligence. Machine learning is one of those ways.

Nested under machine learning is the emerging field of deep learning, which Featurespace used to build Automated Deep Behavioral Networks. The key point to remember is that machine learning is the artificial intelligence technology most commonly used in the fight against money laundering.

How does machine learning work?

Featurespace’s ARIC™ Risk Hub is a real-world example of how machine learning works and its application to preventing financial crimes.

ARIC Risk Hub is built with a machine learning model that studies the behaviors of banking customers. This is built into its transaction monitoring system, a requirement in most AML regulations around the world.

The model has an algorithm that analyses massive sets of transaction and behavior data. It looks at what types of transactions get made, their monetary value, who the recipients are, frequency, time of day and numerous other data points. Collectively, that information gets used to build a profile of each and every customer.

From there, the model can begin to recognize what behaviors are normal for a given customer, and what behaviors are out of the ordinary.

It’s this model that gives financial organizations a full picture of their customers’ activities. That way, an individual customer or an individual transaction can get flagged with precision for further investigation. The machine learning model learns from customer data continuously and in real time, too, so that the alerts get more and more accurate.

To learn more, have a look at this video:

How Featurespace uses machine learning for AML

Financial organizations use ARIC Risk Hub to:

  • Monitor transactions at scale.
  • Prioritize alerts.
  • Reduce false positives.
  • Surface unknown threats by finding new connections in the data.

These are some of the capabilities HSBC was looking for in its insurance business when the company reached out to Featurespace. Our team built a cloud-based implementation of ARIC Risk Hub for HSBC that used rules and machine learning models to monitor transactions and generate alerts.

That implementation has allowed HSBC and its AML team to prioritize alerts more efficiently and effectively. To learn more, have a look at our HSBC video case study.

Get in touch

Featurespace’s machine learning models are built to help organizations detect and prevent financial crime. Our ARIC Risk Hub monitors real-time customer data for anomalous behaviors and helps analysts prioritize alerts.

To find out how ARIC Risk Hub’s machine learning models could support your own organization’s AML systems and compliance, book your demo today.