Introduction

Money laundering underpins and enables most forms of organized crime, allowing criminal groups to further their operations and conceal illicit capital. Technology adoption and the subsequent integration of financial markets has increased the accessibility of international bank transfers, enabling bad actors to move capital globally, at speed, throughout the layering process. At present, anti money laundering (AML) regulations are localized and global minimum standards are yet to emerge. Organized crime groups (OCGs) are able to capitalize on the existing regulatory discrepancies to transfer illicit funds through jurisdictions with less stringent AML supervision.

Machine learning (ML) solutions enhance institutions’ ability to identify and mitigate financial crime by facilitating the analysis of more data with greater accuracy. Global regulators have acknowledged[1] the benefits of such software and almost all banks have ramped up their technology spend in financial crime compliance as a result.[2] However, strict data protection regulation often limits the effectiveness of these solutions as they are only able to run over a subset of data within the perimeters of a single institution or subsidiary. This limits banks visibility into broader chains of suspicious activity and allows many OCGs to operate undetected. Banks, regulators and law enforcement remain firmly on the back foot in the fight against financial crime, and with the value of money laundered each year estimated to be between 2 – 5% of global GDP, [3] a new approach is warranted.

Machine learning solutions within AML

Current financial crime technology solutions leverage ML to enhance money laundering surveillance through the following primary use-cases:

  • Know your Customer (KYC) and Customer Due Diligence (CDD): ML benefits AML activities through increasing the efficiency and effectiveness of customer verification and screening. This enhances banks’ KYC and CDD procedures which minimizes risk and better uncovers potential suspicious activity.
  • Transaction Monitoring: Traditional, rules-based transaction monitoring systems generate suspicious activity alerts resulting from abnormal transaction behaviors. However, it is estimated that over 95% of these alerts are false positives, with only ~2% of alerts resulting in suspicious activity reports (SARs).[4] ML systems support financial institutions in reducing the number of false positives whilst enabling banks to better classify alerts based on their risk levels.

The data sharing challenge

Banks, and the AML software vendors on which they rely, are limited by data privacy regulations which prevent cross-entity and cross-border information sharing. Whilst regulations such as GDPR are vital as a means of protecting the personally identifiable information of customers and subsequently reducing the risk of data leaks, they limit firms’ visibility into suspicious activity across a broader network of institutions. OCGs capitalize on such limitations and employ a range of methods in order to navigate AML controls.

“Smurfing”

Smurfing is a colloquial term for a form of monetary structuring in which criminals deposit illicit capital into multiple bank accounts simultaneously. These cash transactions are spread across different accounts in order to keep them under the $10,000 regulatory reporting limit, so are unlikely to raise alerts within individual banks.

Transactions such as these appear ordinary when observed in isolation. It is only possible to expose larger networks and identify such activity through access to more KYC or transaction data. Data protection regulation hinders institutions’ ability to observe a full chain of transactions across multiple entities and therefore limits the effectiveness of ML solutions in AML.

Emerging data sharing models, are we there yet?

Data sharing, or a lack thereof, exists as the achilles heel of current ML technologies. Private information partnerships are emerging across the globe to facilitate the exchange of financial crime data – examples include Invidem in the Nordics[5] and Transaction Monitoring Netherlands.

Transaction Monitoring Netherlands (TMNL)

TMNL is first-of-its-kind a mechanism tasked with identifying suspicious behavior across multiple banking institutions. In July, 2022, ABN AMRO, ING, Rabobank, De Volksbank, and Triodos Bank joined forces in order to monitor transactions across the entire network.[6] This model sees the implementation of an independently managed, centralized database containing KYC and transaction data from all five banks. By pooling their data, TMNL analysis will draw links between corporates that move illicit capital between institutions to avoid detection, distributing independent multi-bank alerts as a result.

Models such as TMNL are not without their own limitations, however, as sharing masses of information in a central unit is incredibly challenging. Irrespective of regulatory hurdles, institutional distrust remains a barrier for information exchange and encouraging firms to share proprietary and sensitive data with a third-party is no mean feat. As a result, such an approach may struggle to achieve industry-wide buy-in.

Federated Machine Learning (ML), the logical step forward

The Financial Crimes Enforcement Network (FinCEN) and the Financial Conduct Authority (FCA) understand that current AML processes need radical reform. To this end, both regulators have collaborated in launching a set of prize challenges to sponsor the development of “privacy-preserving federated learning solutions”.[7]

Federated learning (FL) is a form of privacy-preserving technology in which an algorithm is trained across heterogeneous datasets. By training the model across multiple entities or servers holding local data samples, FL removes the need to exchange underlying data, allowing firms to achieve the benefits of a collaborative and networked approach, without the risk of sharing sensitive information.

Consilient: leading the charge

Both the U.K and U.S acknowledge the potential of FL in the fight against financial crime and Consilient is well placed to further the agenda, pioneering the use of FL within AML. By bringing the analytics to the data, Consilient has the potential to provide holistic insights from entire financial markets, allowing for far broader analysis than institutions would be able to obtain independently.

While the ML models underpinning existing AML tools are robust and intelligent, they are restricted by the datasets on which they are trained. Consilient’s FL solution offers an opportunity for institutions to overcome the challenges associated with data protection and privacy while fulfilling their desire to analyze data holistically across the industry. As a result, Consilient offers a foundational component for all organizations around the world to utilize as a part of their ecosystem in fighting financial crime.

[1] https://www.fatf-gafi.org/media/fatf/documents/reports/Opportunities-Challenges-of-New-Technologies-for-AML-CFT.pdf

[2] https://www.sas.com/content/sascom/en_us/offers/21q2/acceleration-through-adversity.html

[3] https://www.unodc.org/unodc/en/money-laundering/overview.html

[4] https://www.reuters.com/article/bc-finreg-laundering-detecting-idUSKCN1GP2NV

[5] https://invidem.com/

[6] https://www2.deloitte.com/nl/nl/pages/financial-services/articles/5-dutch-banks-to-make-an-impact-with-transaction-monitoring-netherlands-tmnl.html

[7] ​​https://www.whitehouse.gov/ostp/news-updates/2022/07/20/u-s-and-u-k-launch-innovation-prize-challenges-in-privacy-enhancing-technologies-to-tackle-financial-crime-and-public-health-emergencies/