Alerts optimization and reduction in false positives are essential requirements for a sound anti-money laundering (AML) program. Financial institutions have been so far seen working on optimizing their existing AML platform or tools based on simple analytical formulae that rely on qualitative and expert judgments. As a result, existing solutions generate a significant amount of false positives. This erodes the effectiveness and efficiency of an AML program, by escalating the need for additional trained resources, increasing the probability of missing a valid alert in the deluge of false positives, and eventually pushing up the cost of compliance.
Thomson Reuters’ Cost of Compliance 2017 report indicates that for the first time in six years, the largest percentage of firms (40%) have reported that their AML budget is staying flat. Also, future shortages of trained staff and outdated technology have been stated as the key AML challenges in the Dow Jones & ACAMS Global AML Survey Results 2016. These are prompting banks and financial institutions to review their present AML systems.
So far, the financial industry has been slow in adopting predictive modeling solutions partly because these models were difficult to explain and validate to satisfy regulatory requirements. Moreover, these systems lacked self-learning capability, were incompatible with the existing AML solutions at most banks, and were costly to implement and took a long time to deploy.
Given the above challenges, banks have started to look for simpler solutions that are easier to integrate, compliant with regulations, and do not escalate operational cost.
Analytics coupled with predictive models and machine learning capabilities can help banks raise their game with respect to their AML programs, making them a lot more effective and efficient. We believe that adaptive and predictive statistical models (such as machine learning-based risk scoring and alert-generation engines) can help banks reduce the false positives by about 35%, and also help in driving down the operational cost by close to 30%. At the same time, these solutions work as ancillary software solutions, which can work in parallel to a bank’s existing AML platform.
These dynamically adaptive, automated, scalable, predictive modeling solutions have a combination of multiple individual machine learning models that capture various data nuances and deliver output with a high degree of accuracy. Since these are essentially ancillary software solutions, they are easy to integrate (seamlessly connect to upstream and downstream systems) and can be deployed in the production environment in a short period of time (say, 10-12 weeks), resulting in faster time to market. These tools have the ability to combine data sets from multiple sources, alert data can be imported from connectors and written to pull-push data from multiple databases and technologies. They also come equipped with self-learning capability that enables auto-updation of models with revisions in data (comprises new patterns and feedback from analysts w.r.t alerts). Such tools incorporate human feedback and new data revisions through using machine learning algorithms and AI techniques to continuously detect new patterns and not miss out on any true positives. These solutions have the ability to explain model output in detail, offer robust audit trail (detailed record of model output), and are available for model assurance and validation purposes.
We believe that an analytics-led approach to reduce false positives will help in optimizing the cost and time spent on AML investigation. These solutions would not have any tight coupling or hard dependency on existing AML platforms and can work as ancillary software. Since these type of solutions can work in conjunction with existing AML solutions, no modifications or adjustments are required to the core platform. These solutions will offer robust audit trial and assurance framework, and can therefore meet strict resilience and control requirements. Implementing these solutions however may not be an easy task. What in your view might be the critical success factors in this regard? Please share your thoughts in the comments section below.