Challenges

• Large data volume – Millions of active customers transacting on a daily basis posed a technical challenge to existing simulations.
• Initial threshold set-up – the existing solution was based on statistical “gradient” methods; upon revision it the method lacked the business context of the associated risk.
• Highly granular segmentation (100+ segments) – having such granular customer segmentation required the bank to invest multiple resources to correctly assign thresholds and scores per each population group.

Project Solution

For tuning purposes, the detection scenarios were simulated by recreating them on a test system so they could run quickly and as many times as required by the analysts. In this case, the usage of simulated scenarios also allowed the bank to handle a much larger volume of data.

To further improve the capabilities of the legacy system, a custom scoring layer was created to provide the analysts with the ability to focus on the most important scenarios and riskier clients and activities. This custom layer consisted of assigning a level of priority to each scenario and determining the risk level of each client and account based on the institution’s risk model. A different score was then defined for the resulting risk levels.

With the custom scoring layer in place, all the scenarios were run to detect all potential suspicious events. Those events were then aggregated into cases and only the cases having an aggregated score above a pre-defined cutting score were considered for analysis. The other alerts were filtered out as they were deemed to be low risk.

Compared with production, this approach resulted in a significantly larger proportion (10 times more in the first tuning cycle alone) of cases being generated for riskier clients and activities. Further analysis of the resulting cases also identified scenarios that were strongly correlated and that were often generating events together. This allowed the analysts to identify risks areas that were covered by multiple detection scenarios and provided recommendations to reduce the priority level of some scenarios or even disable those scenarios.

Matrix-IFS Value

Matrix-IFS succeeded to significantly improve the efficiency of the bank’s legacy Transaction Monitoring system by implementing a custom risk-based approach to lower the client’s risk exposure. Instead of replacing its legacy system, Matrix-IFS helped the client to save considerable costs by improving its actual system and extending its lifetime.

Before Matrix-IFS intervention, 99% of cases in the production system were assigned to low-risk clients, thus generating manual work and de-focusing the compliance team from risky behaviour. After the project, 10% of cases were assigned to ‘High Risk’ client types (based on the bank’s classification terminology). The change in distribution enabled the compliance team to focus their investigation resources on cases with a higher level of interest from a risk perspective.