If working with an Anti-Money Laundering (AML) model, it is important to note that tuning is not a one-time effort. This action should occur on a regular basis. As factors like the economy and a bank’s client base evolve, so does the pattern of transactions being monitored by AML systems. It is recommended to monitor and re-tune thresholds every 12 to 24 months depending on how dynamic conditions are.
There are two important components of the lifecycle of suspicious activity models.
The first is the ongoing performance monitoring of a model. Metrics such as the ratio of productive to false positive alerts and historical distribution of alerts help us determine whether a model is performing as expected. In our experience, regulators have cited the lack of performance monitoring as one of the shortfalls of the suspicious activity monitoring system.
A simple dashboard can be built in data visualization software such as Tableau to enable the high-level tracking of model efficiency, in addition to allowing drill downs by rule, customer, etc. Similar views can help both the first and the second lines of defense to track model performance.
The second component of the suspicious activity model lifecycle is the case for automation in the periodic re-tuning of model thresholds. The introduction of robotic process automation (RPA) can substantially reduce the manual effort employed in re-tuning models. Although the qualitative aspect of the tuning process cannot be automated, RPA platforms such as UiPath can help alleviate the labor of the quantitative steps in this repetitive process.
A well-designed bot can be trained to perform recurring tasks such as aggregation of data, and application of analytical techniques including drawing up population distributions in above-the-line and below-the-line testing. An average financial institution spends about $60 million per year on AML-related tasks. Consensus across the banking sector suggests that automation can lead to cost savings of 35-45% in areas where it is applied.
It is imperative for financial institutions to invest in strong AML programs to avoid being penalized by regulators. As we have seen, controls over data quality and application of analytics to model tuning can help banks develop and tune robust models. Additionally, use of data visualization dashboards enable banks to monitor the performance of these models. The advancement of RPA software has given them the opportunity to further the costs associated with recurring tasks and to improve the overall efficiency of their AML programs.