From traditional methods to advanced AI technologies, anti-money laundering (AML) compliance has experienced significant transformation.
According to Consilient, traditional approaches, reliant on (heuristic) rules-based transaction monitoring, often generate a plethora of false positives by flagging legitimate transactions that simply meet certain predefined criteria, such as transfers exceeding or are close to $10,000.
These systems struggle with adaptability and often fail to detect sophisticated laundering techniques that deviate slightly from these rigid parameters. As a result, financial institutions grapple with a barrage of alerts, many of which prove to be inconsequential, burdening analysts with manual reviews and leading to inefficiencies and potential for error.
The financial sector processes millions of transactions daily, challenging the differentiation between legitimate and suspicious activities in real time. Historically, U.S. financial institutions have spent roughly $180m annually on analyst salaries just for preparing Suspicious Activity Reports (SARs), indicating a costly and slow process fraught with human error. It is widely reported that transaction monitoring systems generate up to 97% false alerts and that organizations have lived with this problem for so many years. It is inconceivable for any company in any industry to sustain a 97% product failure rate and not seek change.
To combat these limitations, larger financial services companies have began leveraging machine learning (ML) technologies. Early local or in-house ML models marked a advance in AML compliance, providing an improved efficiency of up to 35%. These models could analyze extensive data sets, identify complex patterns, and adjust to new information more effectively than their predecessors.
However, they were not without drawbacks. Firstly, SARs are a rare occurrence and therefore building robust models without sufficient ground truth (outcomes) is challenging. Secondly, operating in data silos within individual institutions, these models had limited visibility into broader patterns of financial crime across different organizations. Moreover, issues such as biased training data and model overfitting hindered their effectiveness, leading to high rates of false positives and low detection rates for new criminal tactics.
Data sharing is a proven solution to many problems of isolated ‘silo’ working. It ihas been used for credit assessment and fraud for many years to great success. Sharing of intelligence is known to be vital in preventing and solving criminal activity. The challenge for AML is that the data is so highly confidential its security is paramount. The risks of data breaches currently outweigh the value in sharing.
However, there is an answer where sharing of suspicious behaviour and patterns can take place without actual data ever being shared.
The latest development in this technological evolution is federated learning, which addresses many of the shortcomings of the current environment and earlier models. Unlike centralized machine learning, where data is moved to the computation, federated learning brings the computation to the data. This innovative approach allows for the sharing of behavioral insights without compromising data privacy. Financial institutions can now benefit from a collective knowledge base, creating more robust, accurate models for detecting financial crime while maintaining stringent privacy standards.
Consilient’s federated learning solutions demonstrate a marked improvement in efficiency, in one test reducing the number of transactions alerts by 88% without losing any of the originally detected suspicious one. With another organization being able to reduce alerts by 77% and increase suspicious activity discovery by over 50%. This enormous step forward not only streamlines the AML process but also enhances the accuracy of detections.
By learning from multiple independent data sets, federated learning helps financial institutions significantly reduce false positives, lower operational costs, and improve the detection of criminal behaviors using insights pooled from a broader data spectrum. It also enables the identification of new risks that might go unnoticed by single organizations.
Federated learning does more than just improve efficiency and accuracy; it fosters standardized risk management across diverse organizational structures and geographies, providing regulators with confidence that organizations large and small can implement a solution that is fully optimized to prevent financial crime.
Financial institutions that embrace these technological advancements in AML compliance are better equipped to detect and prevent financial crimes, reduce regulatory exposure, and enhance overall operational efficiency. By transitioning from outdated TM systems, manual processes to sophisticated AI-based systems, the financial industry can stay ahead in the ongoing battle against money laundering.
