This article is by Christian de Chenu and originally appeared on the DataRobot Blog here: https://blog.datarobot.com/fighting-financial-crime-with-data-analytics
DataRobot recently participated in the Financial Conduct Authority (FCA) Global AML and Financial Crime TechSprint. The FCA is the financial regulatory body in the United Kingdom, and the event took place at their headquarters in London. DataRobot was represented by myself, a Data Scientist, and André Balleyguier, Chief Data Scientist EMEA, in the competition as part of the "Team Citadel." Our team was led by Karan Jain (Westpac) and composed of individuals from various other companies: Westpac, Citi, Privitar, Oracle, Bureau van Dijk, the FCA, and Companies House.
Ten teams with more than 100 participants competed in the event, using various technologies to positively impact the prevention and detection of financial crime. I am proud to announce that Team Citadel took the first prize in this year's highly competitive event.
Team Citadel. Image courtesy of http://liveillustration.co.uk/
Our goal was to advance the state-of-the-art techniques for anti-money laundering using data analytics and technology to make it more efficient for banks to share data and identify suspicious activities.
Our prototype solution, named Citadel, was developed to help address the EU’s pending statutory 5th Anti-Money Laundering Directive ("5AMLD") which kicks in January 2020.
This solution centered around a new ‘discrepancy reporting requirement’ contained in this directive which will require entities, such as banks, to report any discrepancies they find between the information that they hold, and the information on the central register of companies. In the case of the United Kingdom, the register is called "Companies House" and during the event, we were given synthetic data from this register, as well as data from six banks, designed to be as accurate as possible to real-world data.
We created a process for banks to reconcile their data about the Ultimate Beneficial Owner (UBO) of their corporate clients, eventually allowing banks to be more efficient at complying with regulatory reporting requirements and identifying suspicious individuals and networks related to their customers.
UBOs and Money Laundering
By definition, money laundering conceals the origins of money obtained illegally by passing it through banking transfers and commercial transactions. Companies are at the heart of most money laundering scenarios.
By reconciling the UBOs across a number of banks, the centralized register can correctly assess risk profile information from multiple sources across networks of UBOs and companies and in turn, can be better at identifying other network participants for investigation.
Usually, banks don't know much about where the UBOs related to their entities operate in the wider ecosystem. As a consequence they are not able to capture key risk factors that would allow them to flag their entities as suspicious.
By equipping them with more accurate information on the relationship of various companies via the UBOs, banks can, in turn, become better at predicting suspicious activity.
Using DataRobot to Predict Suspicious Activity
Anti-money laundering subject matter experts (SMEs) all know that suspicious activity often happens within specific typologies of entities, for example, entities whose UBOs are connected to many other entities. To prove this point, we built a powerful machine learning model in DataRobot that would help investigators focus their time on the entities that are likely to be suspicious. We did this by looking at all the entities in the bank's history and observed which ones eventually were reported as having suspicious activity. Learning the patterns in this data allows us to identify which entities are at risk to be flagged after investigation. This model could be used to help prioritize investigation teams to focus only on the right alerts.
After we enriched the data with the UBO profile, we noticed that the most powerful information was now coming from data that captures network and risk information coming from other banks and from Companies House. The bottom line is that better reconciliation => more data about UBOs => better models => finding more suspicious cases -- all with fewer resources. This not only allows more efficient reconciliation but also dramatically helps make the investigation of cases more efficient. Some banks have hundreds of employees focused on manual investigation of cases for anti-money laundering compliance.