ML in AML: Applying Data Science & AI to Tackle International Financial Crime
By Elena Mesropyan for LTP
Since the borders of AI application across industries have not been discovered yet, let’s look at a matter of fighting sophisticated crime schemes, in particular, money laundering. The economic effects of money laundering are vast and highly destructive since money laundering is a problem not only in the world’s major financial markets and offshore centers, but also for emerging markets – any country integrated into the international financial system is at risk.
Global money laundering transactions are estimated at 2 to 5% of global GDP, or roughly $1–2 trillion annually. Yet, according to the United Nations Office on Drugs and Crime (UNODC), less than 1% of global illicit financial flows are being seized by authorities.
One of the reasons money laundering was able to gain such a drastic scale internationally is because every crime of such soft is unique and consists of absolutely legitimate transactions that do not raise suspicion as separate links of the whole scheme. In addition, those schemes evolve over time and it becomes more difficult to flag individual suspicion activity – every action is sharpened to not raise red flags (meeting the deposit limits, timing, etc.).
Meanwhile, AML programs are sharpened to tackle those schemes with the same logic. Legacy AML policies and programs are not flexible enough to notice micro-flags in massive and highly sophisticated money laundering schemes, which results in a poor performance with skyrocketing costs. Legacy technology generates a mass of false positive alerts, requiring high human resource expenditure to look into every case.
Professionals from Booz Allen Hamilton believe that the future of AML programs will not depend on humans since ML is expected to revolutionize the way that AML programs operate. The company suggests that the future of AML will not be defined by the growing number of analysts investigating of false-positive alerts – an approach that is costly, ineffective and not sustainable for the scale of global crime.
Rather, the future will be driven by the advances of machine learning and other technology which can improve the automatic detection of suspicious activity, making it more difficult for criminal activity to go undetected. In other words, the future of AML programs is leaner, smarter and agiler, supported by sophisticated technology that can detect and respond dynamically as their AML risks evolve, Booz Allen states.
Application of machine learning to AML has certain challenges, though, related to the restricted ability to label data and, hence, limited availability of labeled datasets. To tackle that challenge, industry professionals have been emphasizing the role of unsupervised techniques that may be worth considering, among which are network modeling, clustering, time series on graphs and more.
As explained by experts from Brighterion, cross-industry AI platform, unsupervised learning is learning from unlabeled data, where particularly informative privileged variables or labels do not exist. As a result, the greatest challenge is often to differentiate between what is relevant and what is irrelevant in any particular dataset. And since historical data related to money laundering is scarce and unreliable, it is vital to utilize unsupervised learning technologies which have the ability to gain insight from the data without any prior knowledge of what to look for.
The power of unsupervised learning is in the system’s ability to ingest data from a multitude of sources – having a system flexible enough to accept multiple data points from a variety of sources is essential in tracing the full behavior of the individuals and the money/assets laundered, as stated by Brighterion professionals.
At the end, it’s that invisible micro-activity that makes every scheme possible. The ability to track unusual micro-moments of every client’s behavior in parallel with unusual behavior of other sets of agents in real-time could significantly boost the performance of AML programs in a cost-efficient manner.
National institutions in some parts of the world are already actively working to leverage ML in AML. Australian Transaction Reports and Analysis Center (AUSTRAC), for example, was reported to collaborate with researchers at RMIT University in Melbourne to create a system capable of detecting signs of suspicious activity within huge amounts of data. As reported by Bureau van Dijk, the technology, which was trained using previous analyses of suspected money laundering networks, narrowed down millions of transactions to around 750,000 cases for further analysis by human investigators. The edition also reported that the technology is capable of identifying patterns by viewing transaction histories across groups, rather than focusing on individuals.
First appeared at LTP