How AML analytics can transform financial crime programs

Examples of advanced analytics and machine learning techniques that can strengthen AML model input and output include:

• Data profiling and quality assessment
• Entity resolution and network analysis
• Natural language processing and text analysis
• Logic performance testing
• Alert generation simulation and model re-creation for model assessment
• Advanced customer segmentation
• Predictive risk analytics
• System tuning and optimization

Advanced data analytics and machine learning can make better use of available data to improve AML model output and reduce the number of false positive alerts.

Fewer false positives mean AML specialists spend their time reviewing the alerts most likely to represent truly suspicious activity. And with alert review specialists focusing on the alerts that matter most, the organization’s financial crime prevention program become stronger and more cost-effective.

Banks and other financial services companies have no shortage of transaction and customer data to work with. The problem for financial crime prevention programs lies in using and analyzing that data in a way that successfully improves the quality of AML alerts.

Program leaders who want to apply advanced AML analytics must solve several key challenges:

• Overcoming data quality issues. If data isn’t prepared properly, machines can’t learn from it, and analytics can’t produce accurate and reliable output. Missing, incomplete, duplicate, out-of-date, or corrupted input data can all compromise the results from machine learning and analytics.
• Finding the correct analytics techniques. Different machine learning, analytics, and data visualization techniques work best in different situations. Knowing whether a specific problem would be best solved by statistical analysis or by other techniques is critical for alert output quality and program efficiency.
• Deploying specialists with the right expertise. The success of an advanced AML analytics initiative rests on access to experienced specialists who understand cutting-edge data science and AML regulatory requirements. Such specialists are in high demand.
• Translating AML analytics results into insights. Ultimately, leaders of the organization’s financial crime prevention program should set analytics goals that go beyond reducing false positives. Advanced analytics should yield insights that help leaders within the organization make strategic decisions and adapt the AML program to face evolving risks.

The need for sophisticated analysis doesn’t stop when data gets processed and fed into an AML model for transaction monitoring, list screening, or customer risk rating. AML data specialists must perform deep analysis on the source data for AML models and the output data. Improved source data and properly applied analytics techniques should reduce false positives and improve alert quality. If not, either the model or the analytics approach might require further tuning or adjustment.

One proven and highly effective use case for advanced analytics in a financial crime prevention program involves customer segmentation for AML compliance purposes. Advanced customer segmentation techniques based on data analytics and machine learning can deliver vital efficiency benefits and insights for AML teams and the models they rely on.

Customer segmentation in an AML context means that instead of monitoring an organization’s customer base as an ungrouped, unsorted mass during transaction monitoring, customers can be grouped so that they share specific attributes.

For example, your organization might group customers who tend to share specific transaction behavior patterns – say, businesses versus individuals. From there, you might want to separate the businesses into further groups based on the level of AML risk they present to your organization.

For many organizations, this is about as far as it goes. When segmentation depends on constant human input and judgment, segments tend to remain broad.

Advanced statistical analysis techniques and machine learning algorithms can help define monitoring segments, cluster them together, and align monitoring scenarios to that segmentation – all without manual programming. This technology-powered efficiency can lead to much more detailed segments for AML purposes.

Imagine how the quality of alerts from transaction monitoring systems might benefit from:

• Setting different monitoring thresholds for a cash business, such as a convenience store or car wash, versus a consulting or technology company
• Identifying types of business that routinely take large volumes of automated clearing house payments, such as small businesses or sole proprietorships
• Layering in geography to identify transaction pattern differences for the same customer segment from region to region

With the efficiencies from advanced customer segmentation, segments become much more granular. Organizations can use their existing data to identify, for example, 20 different segments of accounting businesses, where before they had two. And with detailed insight into each segment, they can set transaction monitoring thresholds accordingly to reduce false positive alerts.

As with any AML model, the output of advanced customer segmentation is also an input that can find applications in other areas of your program. A more detailed understanding of your customer can feed into and improve know your customer and enhanced due diligence models.

Once you initiate transformation in an area like customer segmentation, the benefits begin to cascade and prime your financial crime program for transformation in other areas, too.