Can Technology Solve the De-risking Problem?

Under the international regulatory framework for anti-money laundering and countering the financing of terrorism (AML/CFT), banks are assigned significant responsibilities for detecting and preventing illicit financial flows. These responsibilities include performing due diligence on their customers, monitoring accounts and transactions for suspicious activity, and reporting suspicious activities to the government.

The “de-risking” problem

In recent years, regulators have raised their expectations for what counts as adequate AML/CFT compliance. At the same time, they have cracked down on institutions that have fallen short. While arguably necessary, this more stringent enforcement has produced some unintended side effects. In particular, it has put pressure on banks’ ability and willingness to deliver certain types of services, notably correspondent banking services.

Correspondent banking—the provision of financial services by one bank (the correspondent bank) to another bank (the respondent bank)—is vulnerable to illicit finance abuse. A correspondent bank generally does not have a direct relationship with the respondent bank’s customers. Often, the only information it has access to is the originator and beneficiary information contained in the payments messages themselves. Therefore, it can be a challenge for the correspondent bank to properly assess the illicit finance risk that such transactions pose. While regulators have clarified that as a general rule, banks are not expected to know their customers’ customers (KYCC), many correspondent banks nonetheless find these types of risk difficult to manage in a cost-effective way. In addition, correspondent banking has traditionally been a high-volume, low-margin business. 

All of this has resulted in “de-risking” whereby banks have decided that some correspondent banking relationships are unprofitable and have proceeded to terminate them. Recent data from the Financial Stability Board show that the number of correspondent banking relationships, especially between rich and poor countries, has declined since 2001. The IMF has also sounded a warning that cross-border flows are concentrated through fewer correspondent banking relationships, which in turn may increase the cost of financial services to poor countries. 

Technology to the rescue

However, there are two new technologies that may help to solve the problem: big data and machine learning.

Big data

Big data refers to datasets that are high volume, high velocity, and high variety. These datasets necessitate different hardware, software, and analytical solutions than do traditional data sets. Banks generate enormous volumes of data in a wide variety of formats. Big data systems can help banks to analyze these data in order to identify abuse while preserving relationships with trustworthy customers.

Big data systems can help compliance staff organize and make sense of large volumes of information. Banks’ compliance staff can utilize data from a wide variety of internal and external sources such as transactions metadata, open-source information (such as negative news stories), and government information (such as sanctions lists, arrest warrants, and so on). Traditionally, these data have been siloed and consequently hard to access. Big data systems can reduce the time compliance staff spend searching for and consolidating information. These systems are typically paired with advanced analytics engines, such as machine learning algorithms, which can help identify patterns and relationships in the data that might have otherwise gone undetected by human investigators.

Machine learning

Machine learning is a type of artificial intelligence which enables computers to learn without being explicitly programmed. There are two broad types of machine learning—supervised learning and unsupervised learning. With supervised learning, the machine learning algorithm analyzes a dataset in order to build a model that predicts a pre-defined output. For example, a supervised machine learning algorithm may be presented with transactions labeled “suspicious” and “not suspicious” and instructed to develop a model that best categorizes transactions as one or the other, based on the available data. With unsupervised learning, the machine learning algorithm explores the features of a dataset, looking for patterns and relationships without attempting to predict a pre-defined output. 

Machine learning algorithms are already being used to tackle money laundering and the financing of terror. The application of machine learning to customer segmentation and transactions monitoring has the potential to greatly reduce both false negatives and false positives in identifying suspicious activities. Clustering, a type of unsupervised learning, can be used to develop much more sophisticated customer typologies than traditional methods. This can help banks to gain a better understanding of their customers’ financial behavior. In addition, classification algorithms, a type of supervised learning, can be used to identify suspicious transactions. These algorithms can be trained so that, over time, their accuracy improves. Recently, HSBC has partnered with Silicon Valley-based artificial intelligence firm Ayasdi to automate some of its compliance functions. Another American company, QuantaVerse, is helping several large international banks to fight money laundering and other financial crimes. 

In early October, we will be discussing the scope of new technologies to address de-risking at RegTech 2017 in Brooklyn, NY. We will also be publishing a report, Technology Innovations to Address De-Risking, in which we will examine this topic in detail. While there have been many positive actions on the regulatory side, our view is that technologies that have emerged over the past few years present very real opportunities to solve the complex problem of de-risking.