Anti-Money Laundering (AML)

Detect Money Laundering in Real-time with TigerGraph

Business Challenge
Traditional Solutions
False Positives
False Negatives
Machine Learning
Business Challenge
Money laundering is a growing problem and has become even harder to track with the proliferation of real-time digital transfers and payments, complex international laws and the rise of cryptocurrencies. The United Nations Office of Drug and Crime estimates that money laundering is 2 - 5% of global GDP, or $800 billion - $2 trillion in current US dollars. Governments across the world have increased regulatory oversight to stop money laundering - 2018 survey from LexisNexis estimated that the cost of regulatory compliance for Anti-Money Laundering (AML) for U. S. financial services firms now exceeds 25.3 Billion US Dollars annually.
Show more
Despite deploying an ever-growing army of AML analysts, banks are paying record fines due to non-compliance. In June 2018, Commonwealth Bank of Australia agreed to pay $700 million in fine for breaches of anti-money laundering and counter- terrorism financing laws that resulted in millions of dollars flowing through to drug importers.
Show less
Traditional Solutions Are Missing the Mark
Virtually all existing AML compliance systems are built upon relational databases, which store information (customer, account, transaction, etc.) in rows and columns. The relational databases are great tools for indexing and searching for data, as well as for supporting transactions and performing basic statistical analysis; however, the relational databases are poorly-equipped to connect the dots and identify hidden relationships among as many as 10+ layers of accounts, which is essential for analyzing money trails and assessing their AML risk. Using relational databases, in order to find potential connections, analysts need to join a number of tables to run queries.
Show more
Such queries could take hours or even days to run, rendering any meaningful analysis of linkages among parties and transactions practically impossible. Detecting money laundering requires going beyond individual account behavior, analyzing relationships among groups of accounts or entities over time often combining information from third party sources. Traditional money laundering solutions built on relational databases were not designed to address this challenge.
Show less
Why TigerGraph, a Native Parallel Graph Database for AML?
Reducing False Positives in AML with Deep Link Analytics
As much as 95% of AML alerts raised with traditional AML solutions are ruled in the end as being unrelated to money laundering. Consider the example, where a new AML alert was raised for a counterparty, Counterparty 2 that has had recent financial transactions with a customer account linked to Suspicious Activity Reports or SARs. Traditional approaches would suggest that this new alert is high risk, because the traditional metrics, such as a high number of alerts generated and multiple SARs filed on the same customer account, all point to the likelihood of high AML risk.
Show more
Nonetheless, through graph analysis, it turns out that of all the previous alerts, only those related to Counterparty 1 became SARs while those related to Counterparty 2 have all been closed. Given that this new alert is related to Counterparty 2, it will likely be more similar with alerts 3 and 4, as opposed to alerts 1 and 2, and thus probably should be closed, or at least marked as low risk. Deep Link Analytics with TigerGraph can identify and group “like” alerts where a conventional transaction monitoring system would miss the relationship.
Show less
Finding False Negatives in AML in Real-time with Deep Link Analytics
Money launderers become more sophisticated every year, creating an intricate network of identities and accounts to funnel their ill-gotten gains. This makes it particularly hard and time-consuming to find the false negatives buried deep inside the mountain of legitimate transactions. Consider the example where a new customer account is added. The new customer account is marked as a low-risk account under a conventional scoring approach, because none of the other attributes measured by the traditional scoring model such as high-risk geography, transaction amount or number of previous alerts or SARs display significant AML risk.
Show more
However, the conventional approach fails to consider the cluster of high-risk customers who have an undeclared relationship with the new customer. The new customer shares a phone number with four other existing customers who have multiple SARs and are therefore designated as high risk. Such hidden or undeclared relationship through a phone number would have been quite difficult to uncover using human review or existing models and systems. Deep Link Analytics with TigerGraph reveals the hidden relationship in real-time and elevates the new alert to a high-risk alert.
Show less
Improving Machine Learning Accuracy for AML
Machine learning has been integrated by multiple financial service providers with their existing AML solutions. However, the accuracy for AML detection is quite poor, as the traditional solutions generate a lot of false positive alerts, feeding those as training data into the machine learning. This, in turn, affects the accuracy for predicting AML activity with AI/machine learning systems. Consider three alerts raised in the example. Let’s look at how the alerts were rated based on traditional features based on account history and how they are classified based on graph-based relationship features computed by TigerGraph.
Show more
Training data from traditional AML solutions for machine learning includes features such as financial transaction amount, SARs filed for the related account and if the particular account is part of a high-risk geography. Based on these features, an alert is raised for Counterparty 2 as Counterparty 2 has received funds from a customer account that is also doing business with Counterparty 1 and there are SARs filed for Counterparty 1 based on past transaction history. Counterparty 2 is also in a high-risk geography. After reviewing features generated by TigerGraph, the alert for the Counterparty 2 is reduced to low risk – all alerts in the past for this account have been closed and none have converted into SARs. Even though the account is in high-risk geography, it does not share phone, address or any other information with a high-risk account with SARs. Graph based features have essentially ruled out a false positive in the AML alert for Counterparty 2. Let’s consider the alert for the new customer, who is not located in high-risk geography, does not have any alerts or SARs as they do not yet have a history with the financial institution. Traditional AML solution will not flag an alert for the new customer. However, graph-based features dig deeper and find that the new account shares a phone number with several customers with SARs: graph-based solution creates a new AML alert that was missed by traditional AML solution and marks it as high risk for further monitoring and investigation. Graph based solution finds the false negatives i.e. AML alerts missed by traditional solutions.
Show less
Getting started with TigerGraph