According to Reuters
, global spending on anti-money laundering (AML) processes is estimated to be approximately $8 billion annually. Growing at a compounded annual growth rate of nearly 187% since 2007, the increase is primarily attributable to unidentified money laundering activity, rapidly escalating fines, and false negatives not picked up by transaction monitoring systems (TMS).
TMS used to be banking’s go-to solution for AML compliance, but now TMS are actually starting to become part of the problem. TMS platforms are far from perfect, generating either false negatives for AML (thus, exposing the firm to unidentified money laundering activity), or a significant amount of false positive alerts that cause compliance costs to skyrocket.
To avoid false negatives, financial institutions have added additional scenarios and relaxed rule thresholds for TMS. However, this causes an avalanche of positive alerts, which requires banks to hire hundreds of analysts to evaluate them. Not surprisingly, the vast majority of these new alerts (as many as 95%) are false positives
and do not lead to SAR filings.
Chief compliance officers (CCOs) and Chief financial officers (CFOs) face a critical challenge: How can they balance the increasing regulatory compliance requirements (and the inevitable hefty fines that come with non-compliance) with the limitations of budgeting for AML program spending? Furthermore, endless tweaking of TMS platforms will never solve this problem. Instead, what’s needed is a new, technology-driven approach to AML.
Three technologies can revolutionize AML compliance. Advances in big data infrastructure and the applications of artificial intelligence (AI), machine learning (ML), and natural language processing (NLP) make it possible to create new AML compliance systems built around a more flexible data schema. These new compliance systems can solve the problems posed by the technical limitations of legacy TMS platforms, such as avoiding false negatives and reducing false positives. They will dramatically improve AML compliance cost efficiency.
The promise of artificial intelligence and natural language processing for anti-money laundering
While artificial intelligence (AI) has been around as an academic discipline since the time of Turing machines, AI is just now coming into its own as a disruptive force in many areas of analysis where ‘heuristic,’ or imprecise pieces of information, can be put together to discover the full answer to the jigsaw puzzle in a client’s account information.
The most complex problems in the AML world involve finding relationships between disparate sets of transactions. These relationships can reveal themselves in a precise way e.g., dollar amounts transferred between the numbers of parties in a network where the network actors’ identification is fairly standard and easily parsed (e.g. Fairchild Co, Fairchild Co. Ltd, or Fairchild).
In this space, machine learning (ML) techniques can find trends or activity pairings to reveal hidden ownership structures. Similarly, ML techniques can perform outlier analysis for anomalous transactions and network analyses for the discovery of networks of seemingly unrelated actors.
Deploying a series of ML detection models decreases the window of exposure for new services, products and acquisitions.
ML enables AI through various algorithms that make it possible to decipher activity between actors that may be anomalous and possibly suspicious. These actors may range from simple structuring or repetitive movement of funds activity, designed to stay below the radar, to rapid movement of funds between accounts where the algorithmic scan reveals the account’s ultimate beneficiary.
ML also solves the problems of data integration and compliance that arise when a new product is launched or when a company is acquired and the acquiring bank needs to ensure that the new service is compliant. Integration effort is low with ML, even though the new service or product will have different data fields than those of the acquiring bank. While traditional TMS platforms, with their rule-based models look for particular scenarios and demand specific fields to be structured in ways that map them to the internal data model, ML does not require a strict data model. Rather, ML simply works with the data that it is given. Because the data does not have to be transformed to fit into a strict data model, the time for compliance is minimized and fines are avoided.
Natural language processing can solve the customer-due-diligence dilemma
As of May 11, 2018, the Financial Crimes Enforcement Network requires banks to perform customer due diligence (CDD)/beneficial ownership checks for all new accounts. In addition, under the new regulations, renewing accounts are treated as new accounts at their renewal date. These regulations are intended to prevent money laundering through “straw men” beneficiaries for accounts and to ensure the validity of ownership for all accounts. As a result of this new rule, banks may face a compliance challenge since ultimate beneficiaries and beneficial owner information is not always readily available. This is especially true for wire transfers where the parties/counterparties are not the direct customers of the financial institution.
NLP is the key technology that will enable institutions to overcome the challenges of the increasing volume of unstructured data and to meet the demands of growing regulatory pressures to gather the required intelligence in a timely manner.
As one of the newest automation technologies, natural language processing (NLP) can help reveal relationships from unstructured text. In dollar-clearing wires, or any form of payments to international counterparties, identifying ultimate beneficiaries represents a puzzle to be solved through discovering idiomatic meaning behind seemingly innocuous unstructured text. This is especially true in wire transfers between various entities, geographies and actor types, where most of the actual activity is coded in the instructions part of a wire. NLP provides an automated decoding unstructured text, and compare it to structured data to analyze for hidden relationships.
New technologies, new challenges: Recommendations
These new technologies can be transformative for banks that are interested in improving regulatory compliance, maximizing ROI, reducing AML investment and avoiding penalties and fines.
Among progressive banks, the most rapid adoption rate for AI, ML and NLP is for know-your-customer and customer-due-diligence compliance. NLP is used to help refine the value of negative new searches. AI and ML aid in new customer risk scoring. In addition, many fintech firms are coming to market with innovative on-premises and cloud solutions that add options to this space.
A second, less well-developed AML application for AI and ML is for transaction monitoring and anomalous activity detection. While there are a number of fintech firms with impressive offerings in this space, when it comes to adoption, caution is the best course of action for now. Many of the models in the market place will need significant validation before reaching widespread regulatory acceptance. So, these anomaly detection models based on ML should be used only as an adjunct to rule-based TMS.
Finally, the area of anomalous activity analytics should be an area to watch and consider to adopt in the short-term. These analytics tend to be used for trend analysis rather than for individual transaction monitoring, and they reveal high correlation rates between model output and real world activity trends.
While new technologies promise new solutions, they also have their limitations. At this point, these technologies cannot replace existing TMS platforms. Yet, they are powerful enhancements that have the potential to increase detection accuracy and to reduce false positives and investigation efforts drastically, all in a manner that is understood by regulators. And looking ahead, these technologies have the potential to unmask the most sophisticated money laundering schemes on a global scale.
Director, Financial Services Advisory
+1 212 624 5556