arrow left facebook twitter linkedin medium menu play circle

Uncovering Hidden Patterns to Improve AML Detection

By Igor Bidny October 25, 2018

Photo of Igor Bidny

about Igor Bidny
Igor Bidny is the Director of Strategy at DataVisor, Inc. Most recently he oversaw Strategy for Financial Services at Ayasdi, a leading enterprise AI company, and has spent a number of years in the analytics, financial services and fintech space covering capital markets, risk and compliance, and asset management. He is a CFA charter holder and has earned a Masters in Finance from the London School of Economics and an MBA from Columbia University.

Uncovering Hidden Patterns to Improve AML Detection

The rising levels of financial crime and an ultra-stringent regulatory landscape make the compliance function an absolute necessity for the AML and compliance leaders. Evolving money laundering patterns are leading to huge fines and mounting pressure on FIs to become more vigilant. While criminals are on top of their game, financial institutions also have the tools and technologies to safeguard their interests by augmenting their current Transaction Monitoring Systems (TMS) with AI and machine learning approaches.

In this piece, let us examine the value that unsupervised machine learning (UML) brings to the table. The challenge of AML teams is two-fold. Firstly, 50% of suspicious activities are undetected by the current TMS resulting in fines and sanctions from regulators due to false negatives. Also, lookbacks are dreaded since they involve a lot of time, money and effort. Secondly, 95 percent of alerts generated are false positives resulting in wasted time spent on investigating them without producing SAR filings. Herein lies the unique ability of the UML technology that can look at massive quantities of data across accounts, customers and time, to link related activity together through shared attributes. By doing so, UML can detect hidden networks of money laundering activity. Unlike supervised machine learning, unsupervised machine learning finds networks and clusters of money laundering to uncover hidden links between millions of accounts. It’s ability to provide a global view to identify suspicious activity and reduce false negatives is unparalleled. So how does it come together?

Inherent Merits of Unsupervised Machine Learning 

Unsupervised Machine Learning (UML) is built in a way that it does not require historic labels or training data and looks for unknown patterns and results. Since it does not know what it is looking for, it has no predetermined bias or pre-set rules. Because UML looks at shared attributes across all transactions, customers and accounts instead of comparing activity to known scenarios, this approach detects suspicious activity without prior knowledge of exactly what a money laundering scheme looks like. This aspect is a huge differentiator from supervised machine learning and rule-based models, which require knowledge of previous patterns to catch similar ones in the future. Using UML to perform global network detection reveals money laundering patterns much more holistically. Its proactive nature is a big plus as it can often provide a significant lift to detection results over existing TMS.

Secondly, this approach provides correlation analysis in real-time. It enables processing events and account activities to analyze the correlations and similarities across millions or hundreds of millions of accounts in real time. Using a distributing computing framework such as Apache Spark allows UML to be applied to large sets of data fields to look at all accounts at once. While other approaches look at accounts in isolation, the algorithm here is set up to show hidden structures across money laundering accounts in real-time. For example, seeing sixty accounts linked together doing similar unusual activity is much more suspicious than seeing one account doing the same strange activity. By finding the entire network together, any alerts created will have a much higher likelihood of being truly suspicious. Moreover, a good quality of alerts reduces operational costs. The analysis of a global population of accounts simultaneously reveals subtle correlations amongst them helping AML leaders make timely decisions.

Thirdly, new sophisticated attacks often involve many different types of events making fast and effective manual rule derivation or updates impossible. The UML approach doesn’t need frequent retuning. Its predictive power is not based on intelligence derived from historical experience as it adapts constantly to the evolving and unfolding suspicious activity patterns. This ability to self-tune is a huge strength.

Powering Up

According to Gartner, by 2021, 50% of enterprises will have added unsupervised machine learning to their suspicious activity detection solution suites.  DataVisor’s technology incorporated into AML solutions offers significant benefits, including transparency, a reduction in compliance costs and improvements in the transaction monitoring process leading to greater overall efficiency.

Popular Posts

Intelligent solutions. Informed decisions. Unrivaled results.

DataVisor Digital Fraud Trends Report: Q1 2020

Learn More

Download the new “State of Sophistication in Fraud” report from DataVisor to ensure your business is equipped to…

Download the new “State of Sophistication in Fraud” report from DataVisor to ensure your business is equipped to prevent high sophistication threat attacks.

A Guide to Centralizing Data Intelligence

Learn More

Discover AI-powered fraud strategies for preventing financial and reputational damage in this powerful e-book.

Online marketplaces withstand a complicated array of fraud attacks—spam, scam, and all points in between. Only the most comprehensive, proactive AI-powered solutions can fully protect against reputational and financial damage. This e-book details the entire lifecycle of a fraud attack, and lays out…

A Practical Guide to Eliminating False Positives in Fraud Management

Learn More

This e-book explores the next generation of fraud prevention technology, which applies unsupervised machine learning to reduce false positives and risk.

This e-book explores the next generation of fraud prevention technology, which applies unsupervised machine learning to reduce false positives and risk.

Protect your business, your customers, and your data.

Request Demo