Money laundering transforms profits from illegal activities—such as fraud, drug and human trafficking, organized crime, and corruption—into seemingly legitimate earnings by concealing the source of the acquired funds.
Recent estimates show that approximately $800 billion to $2 trillion is laundered annually through the global banking system. That’s roughly 2 to 5 percent of global GDP. In recognition of the growing problem, regulators are developing stricter policies—and handing out heftier fines when institutions are caught laundering money.
This is a significant problem for many financial institutions, as those laundering money use increasingly sophisticated methods to evade detection. Although banks are typically on the front lines, other industries used to conceal the source of funds include academia, real estate, hospitality, and healthcare.
Anti-money laundering (AML) mitigates the flow of illegal funds
Traditional AML performed by a bank uses a customer’s profile and transaction history to generate risk ratings and flag various suspicious behavior, such as cash deposits over $10,000. This rules-based approach is critical, but it doesn’t account for funds laundered through networks of individuals in smaller, non-rounded dollar amounts to avoid detection.
As a result, customer-risk rating and transaction monitoring models used by banks often exhibit false positive rates of over 98 percent. Although this evidences a conservative approach that may be appreciated by regulators, it can have the effect of diverting resources away from the highest-risk cases.
In appreciation of these growing challenges in AML, regulators have signaled that they are open to banks developing innovative methods to stay ahead of today’s tech-savvy criminals. Many leading institutions are exploring the use of natural language processing, network analytics, and other machine-learning and AI-based techniques to identify subtle indicators of illicit activity.
Network analytics in the real world
Network analytics examines the connections between related entities to better illuminate relationships. Instead of analyzing an individual, subcomponents of the network are reviewed for similarity to known methods of money laundering and atypical customer behavior.
Networks are formed by links between customers and related activity. These (sometimes inferred) links can be internal data, such as account transfers or joint ownership, or external data, such as a shared address or common use of the same ATM.
Network analytics compliments existing machine learning and fuzzy logic-based approaches that many banks use for AML monitoring. Network statistics (for example, connectivity) for each customer can be used as an input to improve the accuracy of customer risk rating or transaction monitoring models. Fuzzy logic-based approaches that resolve customer identities can also be improved by looking at how closely accounts are connected. In addition to improving the effectiveness of existing techniques, network analytics provides investigators with new capabilities. For example, community detection algorithms can identify the presence of customer groups that could be indicative of criminal behavior.
For example: The Smith family is laundering money by dividing large transactions into small deposits, filtered through online bill payments into temporary accounts. The payments are then used to purchase a boat which is quickly resold for cash—creating a paper trail to clean the money.
Network analytics can help identify the Smith family’s illegal activities. Here’s how it works:
Step 1: Build the Smith network
Begin with Mrs. Smith and identify all other entities, including accounts and people, that she is connected to.
Step 2: Create connections
Next, add the relationships between the individuals, their respective accounts, and any related activity showing payments made within the system to show the flow of funds.
Step 3: Infer relationships using non-traditional data sources
Use enriched data about individuals and their related accounts in order to uncover inferred connections that show suspicious or anomalous activity that might suggest money laundering.
Network analytics is the future of AML
Network analytics has the potential to significantly improve the effectiveness of AML programs. In practice, statistics from a network (for example, how closely it resembles a known money-laundering typology) would be incorporated into existing customer-risk rating and transaction monitoring models as inputs to improve model accuracy. New capabilities such as community detection would help accelerate investigations and identify hidden risks.
Network analytics takes time to get right and can require an enormous amount of computational power to sift through all current and past customer relationships. Historically, uniquely identifying a customer across systems to build links was also quite difficult. But this has changed over the past three to five years as banks have invested heavily in data infrastructure and built unique customer identifiers that are shared across systems. Scalable infrastructure (for example, Hadoop, AWS) has also provided institutions with more storage and computational power—enabling new use cases including network analytics.
Start by building a network of existing customer links by using account transfers, shared account ownership, and payments to build linkages both internally and to external institutions using the destination account number. Then create inferred links between customers by looking at shared addresses, employer, or social media data. Although often the target state, an enterprise grade graph database is usually not required—data can be stored in a standard relational database to get started. Even without advanced analytics, creating this database of links will accelerate investigations and provide data scientists with a rich asset that can be used for AML, in addition to a wide variety of other use cases (for example, marketing).
To take full advantage, most institutions will need to build capabilities in network science as the tools may be unfamiliar to even experienced data scientists. This will unlock a significant opportunity to improve both customer risk rating and transaction monitoring. The secrets to success are having the right external data sources and network science capabilities, and using deep subject matter expertise to inform model development.