For years, financial institutions have been using automated monitoring procedures to make their processes for detecting money laundering and terrorist financing as effective and efficient as possible. However, many of the solutions used still leave much to be desired in terms of effectiveness and efficiency. Silo architectures, a lack of cross-departmental collaboration and the use of multiple decoupled systems in different areas reduce the success of monitoring. Anti-money laundering (AML) and counter-terrorist financing (CTF) experts regularly complain about insufficient hit rates and, at the same time, an excessively high false positive rate, which necessitates a considerable number of unnecessary manual verification procedures. ‘Unnecessary’, because technologies have long been available that could drastically improve the hit rates of monitoring procedures.
Over 100 billion euros – this was the sum estimated by an ‘unreported crime’ study conducted in 2016 for the volume of annual money laundering in Germany. In the new annual report of the Financial Intelligence Unit (FIU) for 2020, a total of 144,005 Suspicious Activity Reports (SARs) were recorded in Germany, an increase of 25 percent on the previous year. Within the last ten years, the annual volume of SARs has even increased twelvefold, according to the federal government’s anti-money laundering unit. To uncover and combat these huge financial flows that can be traced back to criminal activities, the state essentially relies on tips from the financial sector. 98 percent of all SARs submitted in 2019, according to the latest annual report of the Central Financial Transaction Investigation Unit, originate from the financial sector. The latter is required by law to check payment flows for potential criminal anomalies and report them to the relevant authorities. For years, banks have relied on a combination of digital solutions and AML and CTF experts.
The problem: classic AML/CTF solutions have underperforming hit rates.
Monitoring systems work upstream from the experts, checking the mass of daily financial transactions, sorting out and flagging those that are suspicious. As a result, human specialists can concentrate entirely on the cases with the greatest level of suspicion. Nevertheless, an AML/CTF expert handles an average of more than 300 cases per month. This can become quickly unmanageable, and more than 80 percent of flagged transaction turn out to be a false alarm. Accordingly, the hit rate of most of the solutions used still leaves a lot to be desired. The main reason for this: The quality of the analysis procedures is too poor, and the quantity of the evaluated data is simply too low.
Silo architectures complicate overarching analysis
The solutions in use, such as customer due diligence, watchlist screening and suspicious activity monitoring, are often based on a silo architecture. Their data models and technologies are often incompatible with each other. In addition, they frequently cannot fully access the data sets available to the bank because they are stored in data silos. However, data preparation and availability is only part of the picture. An additional challenge is applying the right technology.
Machine learning techniques implicitly use information from data sets to identify relationships and correlations in data and can process large amounts of data quickly. To do so, they require statistically significant amounts of high-quality and task-specific data. In the financial industry, these are not always available in sufficient quantities due to ever-evolving payment platforms and types, rapidly advancing digitization initiatives, and other innovations. This is where knowledge-based techniques such as fuzzy logic come into play. Knowledge-based AI approaches map expert knowledge explicitly and in a comprehensible way. This provides the necessary explainability when it comes to decision making, but also require a high level of expertise.
When looking and machine and knowledge-based approaches separately, it quickly becomes evident that there are major challenges in each case. Purely machine-based processes are based on the permanent availability of all relevant data, which cannot always be guaranteed. Knowledge-based procedures work best when highly qualified compliance experts with years of training are involved. Together, the respective challenges can be solved, and the disadvantages overcome. It becomes clear that the combination of these two technologies, machine and knowledge-based AI approaches, is the most effective way forward.
Hybrid AI methods bring algorithms and experience together
If so-called hybrid AI methods are now used the hit rate of the analyses can be noticeably increased. Hybrid implies that the analysis and monitoring methods rely on both data-based and knowledge-based AI procedures. The former includes, for example, supervised and unsupervised machine learning, pattern recognition and data mining, the latter mathematical algorithms, fuzzy logic, dynamic profiling, and logic score cards. This combination of data-based AI algorithms and experience and insights gained over years by countless AML/CTF experts represents the most powerful anti-money laundering and counter-terrorist financing analytics option currently available.
Holistic platform approach – AML/CTF monitoring from a single source
If this method is combined with a holistic platform approach, the dilemma of silo architectures can be overcome. In this case, monitoring procedures are not used as isolated applications, but are integrated via a central platform. Data and analysis results from the AML/CTF application areas can be comprehensively interlinked and correlated via the platform’s interfaces. They also allow the integration of a wide variety of data streams. Thus, data from all channels and sources of a financial institution can be integrated into the analysis and monitoring processes. Banks gain a comprehensive overview of potential criminal activity and can significantly increase the hit rate of their AML/CTF monitoring.
It is certainly in the interest of virtually every bank to make their AML/CTF monitoring procedures work in a compliant manner with high effectiveness and efficiency and to radically reduce the number of personnel-intensive review procedures for their experts. To achieve this, they will not be able to avoid implementing a solution with a holistic platform approach and hybrid AI technology.
