Financial crime is fast-paced, with bad actors relentlessly seeking gaps and weaknesses within the structures of the global financial system to exploit and circumvent.
As financial institutions and formal players in this space have continued to integrate new technologies into their processes, so too have these nefarious groups to make their tactics more sophisticated and harder to detect.
In response, regulatory bodies and financial institutions are constantly working to adapt and improve anti-money laundering initiatives. For financial institutions, operating systems are at the heart of effective countermeasures and robust compliance, and the most effective AML models are those that are regularly tuned and optimized.
Tuning and optimizing involves adjusting an AML model's configurations to improve performance.
Tuning specifically involves adjusting the parameters of an existing system to improve its performance without changing its fundamental structure. This includes making better use of system resources, reducing bottlenecks, and increasing throughput.
Optimizing, on the other hand, involves more fundamental changes to the system's design or algorithms to improve performance. This can include restructuring code, adopting more efficient algorithms, or changing the underlying technology.
Both exercises are the bedrock of ongoing monitoring, one of the pillars of model validation (more on this below). To meet US regulatory requirements for model validation, AML models must be tuned regularly.
Tuning and optimization always begins with an analysis of model output.
In the case of transaction monitoring systems, this means analyzing the model's alert output by rule and segment to see what's driving alert volumes. It also involves analyzing the productivity of each rule and segment to understand the percentage of alerts and cases that ultimately turn into SARs.
There are numerous indicators that tuning may be required, including:
There are also metrics for sanctions and customer due diligence models that can be used to track the health of the model or indicate that tuning might be required. Industry best practice is to track these types of metrics on an ongoing basis so your institution is always aware of the model's performance.
Some of the sought-after improvements in performance include effectiveness, efficiency, and adaptability.
Customer risk will be more accurately captured and reflected, enhancing risk detection. False positives and negatives will be reduced allowing FIUs more time and resources to focus on the alerts and investigations most likely to yield meaningful results, improving operational efficiency. Finally, the system will be adaptable to emerging threats and regulatory changes and will not degrade over time.
When these improvements are combined, and the model runs smoothly and generates the optimum output, it can positively impact and improve the overall compliance program.
Regulators require banks and financial institutions to prove their systems have robust governance, efficient risk protocols, and sufficient internal control mechanisms capable of managing financial crime risk; accurate data and reasonable filtering criteria are fundamental to this.
Likewise, regulatory bodies, from the Financial Crimes Enforcement Network (FinCEN) in the US to the Financial Conduct Authority (FCA) in the UK, often require financial institutions to maintain effective transaction monitoring systems.
While there is no specified cadence for tuning and optimizing, setting effective AML transaction monitoring thresholds should be considered an ongoing process; your examiners, auditors, and regulators will expect this as part of a risk-based approach.
Triggering events such as mergers and acquisitions, new products or services, new geographic segments, as well as evolving customer risk profiles, criminal methodologies, new regulatory standards and so on, all warrant scenario analysis to adequately address risk and support compliance.
Model validation and tuning and optimization are separate yet complementary processes in the lifecycle of a predictive model.
Model validation assesses the effectiveness of a model. It involves checking whether the model works as intended and meets its predefined objectives. It includes evaluating the model's predictive accuracy, reliability, and consistency over time. It also assesses the appropriateness of the model structure, inputs, assumptions, and parameters.
While both are important, they serve different purposes: validation ensures the model is correct and reliable, while tuning and optimization focuses primarily on improving performance.
Fine-tuning and optimizing systems are complex, data-intensive tasks. With countless variables at play, not tuning a system correctly or regularly can compromise the health and effectiveness of your compliance program, exposing your institution to even greater risks.
The key to compliance longevity is to do this exercise regularly and to team up with a specialist partner with deep technical and data-specific knowledge who understands the intricacies of system tuning and optimization.
Our experts have a proven track record of success, performing dozens of tuning exercises across a myriad of systems for financial institutions across the globe, and are ready to make your system work harder and smarter for you. Fill out our contact form, and let's start the conversation.
AML Conversations – The Solutions Series: Tuning and Optimization
AML Conversations – The Solutions Series: Tuning and Optimization Part 2
AML Conversations – The Solutions Series: Tuning and Optimization Part 3