World Economic Forum Op Ed
Narcotics and human traffickers and other criminals exploit the global financial system to run and expand their illicit enterprises. It is the responsibility of financial institutions to protect the global financial system from being used by these bad actors, which is no small feat. Financial institutions (FIs) wrestle with determining the “true identity” of customers and continuous monitoring of their transaction activities, compounded by the expense and manpower needed. In fact, anti-money laundering (“AML”) and Know Your Customer (“KYC”) compliance spending was estimated to hit USD 32.1 billion globally in 2019. The cost and effort involved would be justified if successful – unfortunately, the United Nations Office on Drugs and Crime estimates that between $800 billion and $2 trillion is laundered globally every year.
That money comes from a myriad of illicit sources including human trafficking, estimated at $150.2 billion per year generated globally; narcotics trafficking, estimated at $426 billion to $652 billion; and bribery and corruption, estimated at 5% of global GDP or trillions of dollars globally. Narcotics trafficking proceeds first point of entry into the financial system is primarily cash; the laundering of bribery and corruption proceeds often relies on the use of corporate vehicles, trusts, or non-profit entities; and the laundering of human trafficking proceeds relies on various financial products including cash, cryptocurrency, and online payments. Each complicates detection in different ways and each is increasingly impervious to detection using static rules e.g., if there is a round-dollar amount over 10,000 going to a high-risk geography then alert.
So how can FIs improve efforts to stop dirty money? The answer, in part, is that they need better tools. Tools that for instance, can analyze publicly available data obtained from organizations like Traffik Analysis Hub, such as cell phone numbers taken from advertising websites of suspected traffickers, to cross-reference against customer information files.
Here, we discuss four ways that machine intelligence (MI) is being deployed to make the detection and prevention of financial crime both more effective and efficient.
Financial institutions must verify their customers’ identity and conduct due diligence. One step of verifying identity is confirming ultimate beneficial ownership of entities, for which financial institutions rely heavily on the customers’ themselves for this information – making it easier for inaccurate and false information to be entered. FIs must also screen publicly available information for “negative news” on their potential customers, customers, and their customers’ customer. These searches result in a plethora of information that needs to be reviewed – but often does not pertain to the individual that was the subject of the search, costing time and resources. Even for FIs using automated onboarding solutions, the onboarding process can take up to 12 weeks and cost an estimated $6,000 to $25,000 per client.
Natural language processing can be deployed by FIs to locate ownership information in unstructured data alleviating the need to collect this information from the customers themselves. And as more data becomes publicly available, e.g., from leaks like the Panama and Paradise Papers, from public registers, and from NGOs like Liberty Shared that collate information from public registers, the more crucial this technology becomes. At the same time, more data means more sorting. That is where machine intelligence can help. Whether it is sorting through identity and beneficial ownership information or reviewing “hits” during the screening process to isolate relevant information, MI solutions can transform the process. MI techniques can eliminate a manual process that can only access and process a fraction of the available data, with an automated process that gathers and filters data from numerous data sources with accuracy and speed.
FIs are responsible for monitoring transaction activity and identifying suspicious activity. In practice, this means that every transaction is fed through a transaction monitoring ("TM") system set to alert when indicators of potentially suspicious activity are detected e.g., large round dollar transactions being sent to or from a high-risk jurisdiction. Even a perfectly configured TM system, however, produces a large amount of false positive alerts (some estimates put this figure as high as 98%). By using behavioral analytics, an MI model can ingest biographical and transactional information about an FIs customers, and then segment those customers into groups that behave similarly e.g., a segment compromised of mostly retired individuals with low electronic activity who are more often the beneficiary than the originator. By systematically segmenting customers by behavior, detecting anomalous behavior in this group becomes easier – more importantly, suspicious activity alerts tend to represent real risk of potential illicit activity. In one use case, Intelligent Segmentation led to a 45% reduction in false positive alerts and allowed the FI to identify suspicious activity that previously went undetected.
Where intelligent segmentation can reduce the noise and identify more potentially suspicious alerts, entity resolution with network generation can help investigators review those alerts by enabling a more holistic use of the information. These solutions can bring together millions of data points from internal and external data sources, including social media, news feeds, and corporate registries to create a complete picture of a customer and their activities. It can then build a network showing how that customer transacts and interacts with others - revealing otherwise hidden connections and relationships. This is especially useful in areas such as trade finance and financial markets, which may have multiple accounts within the FI, fast moving transactions, multiple record keeping forms, complex organization structures and related transactions occur across different systems and timelines.
Shifting from a rules-based approach to a behavioral network approach as outlined above has reduced false positives dramatically, while also identifying potentially suspicious activity that was otherwise undiscovered.
Besides monitoring for suspicious activity, FIs must conduct screening of their customers and transactions to identify sanctioned individuals or entities. MI, specifically supervised learning, can be used to review alerts that indicate a match to a sanctioned individual or entity, and detect obvious false positives, e.g., Scuba Diving Inc. matching to “Cuba” on the OFAC sanctions list. By removing the obvious false positives, FIs can deploy resources to focus on higher value alerts. One use case resulted in closure of over 40% of false positive matches without human intervention, aside from the required quality assurance reviews.
The tools outlined above can be utilized individually and in combination to address the most difficult challenges facing FIs today, and if deployed correctly, can help FIs detect the proceeds of narcotics and human trafficking and bribery and corruption in their systems. The use of machine intelligence is happening all around us, but there have been few use cases for MI to remedy the fraudulent misdeeds and social injustices of the world – this is changing.
This article was part of the World Economic Forum Annual Meeting and was originally posted on the World Economic Forum website on Jan. 15, 2020.
Special thanks to contributors Rachel Sazanowicz and Brian Karp.