Transforming KYT: The use of AI and machine learning in transaction monitoring

This article was originally published in The Payments Association EU’s June 2024 whitepaper, “KYT Best Practices and More,” written in collaboration with Deloitte and Banking Circle. Download the full whitepaper here. 

Efficient and accurate data analysis is crucial for effective anti-money laundering (AML) programs. However, AML teams using outdated transaction monitoring programs often face backlogged systems, with their analysts frequently experiencing burnout due to processing high volumes of alerts with too many false positives. Without a way to triage incoming alerts, highly qualified investigators can spend most of their working days on repetitive tasks like clearing overloaded systems and low-risk alerts.

This causes frustration and wastes company time, financial and energy resources, overloads personnel, and increases the likelihood of teams missing illicit activity. It can also result in unwanted organizational costs, as burnout leads to high turnover rates and costs to recruit and train replacements. Furthermore, if a company is deemed to have insufficient risk management processes, it may face regulatory fines, legal action, and reputational damage. Combined with online payment fraud losses estimated to exceed $362 billion by 2028, the stakes for financial institutions (FIs) are high. 

Enter artificial intelligence (AI).

It might feel like every conference, webinar, and white paper in the compliance industry is talking about AI and its transformative potential for financial crime risk management. But what does this potential actually look like in relation to enhancing firms’ know your transaction (KYT) protocols? 

This article explores the use of AI and machine learning (ML) for transaction monitoring, highlighting five key benefits that are helping firms increase the effectiveness of their financial crime-fighting efforts. 

Five benefits of AI-driven transaction monitoring systems

1. Adapt to changing behaviors in real-time

Unlike static rule-based systems, AI-driven transaction monitoring systems can learn and adjust in real-time, staying ahead of emerging risks. Machine learning algorithms within these solutions analyze historical transaction data to identify trends associated with legitimate and suspicious activities. As criminals evolve their tactics, the AI model dynamically updates its understanding of these behaviors. It adapts by recognizing new patterns and adjusting risk parameters, ensuring it stays ahead of emerging threats. This dynamic adaptation allows the system to effectively respond to shifts in the financial landscape, promptly identifying anomalies and potential risks.

The benefits for financial institutions (FIs) are profound. First, real-time adaptation significantly reduces false positives as the system becomes more discerning in distinguishing between normal and suspicious activities. Second, it enhances the system’s agility, swiftly recognizing and responding to emerging fraud patterns. This proactive approach mitigates the risk of overlooking sophisticated fraudulent schemes, safeguarding the FI’s assets and reputation.

2. Identify hidden relationships to uncover patterns and connections

In addition to tracking changes in behavior, pattern recognition in AI-driven solutions can expose connections within the intricate network of financial transactions. Using graph-based representation, AI algorithms can analyze transaction nodes and entity links to identify clusters and unusual connections. 

With this information, companies can better detect new or emerging fraud typologies and establish rules to mitigate them. For example, at the beginning of the pandemic, global payments firm Lumon noticed a sudden rise in COVID-related investment fraud. With ComplyAdvantage’s Transaction Monitoring solution, Lumon was able to develop and implement new rule sets within 48 hours to combat the threat and prevent more customers from falling prey to fraudulent activities.

3. Efficiently triage alerts to minimize false positives

When an alert is triggered, AI models can evaluate the risk level based on various factors, including transaction amounts, frequency, and deviations from established patterns. It then assigns a risk score to the alert. Instead of relying on static rules that may generate false positives due to rigid parameters, the AI model dynamically adjusts its understanding of what constitutes suspicious behavior. Continuous learning is a key mechanism in this process. Feedback from analysts, investigations, and outcomes of previous alerts are fed back into the model, allowing it to refine its algorithms and improve accuracy over time.

As a result of efficient triaging, the identification of high-risk activities is accelerated, enabling a more rapid response to potential threats. Additionally, false positives can be reduced, preventing unnecessary investigations and directing the focus towards genuine risks. This streamlined process enhances the effectiveness of transaction monitoring and improves the overall operational efficiency of the FI.

4. Produce deeper insights to meet regulatory expectations

To further boost the confidence of compliance teams when making decisions, AI-driven systems can provide a deeper understanding of the reasons behind alert generation. This level of transparency becomes especially important during audits since the concept of explainability has become a growing area of concern and legislative focus. Regulators increasingly require those who use or provide AI models to provide transparent and traceable decision-making processes, as well as clear and understandable information on the AI model’s capabilities and limitations.

Interestingly, our State of Financial Crime 2024 survey showed how firms are thinking about AI – and the results were, at times, contradictory. While most firms believed they were on track to meet regulatory expectations around AI, 89 percent said they were comfortable trading off explainability to improve efficiency. 

5. Precisely tune rules for more targeted monitoring

Contrary to common misconceptions, machine learning doesn’t replace rules; it complements them. Rules provide the foundational knowledge of customer behavior that machine learning thrives on, forming a symbiotic relationship. Many firms initiate their system with baseline rules, gradually integrating a more sophisticated, data-centric machine learning model. This phased approach allows time for thorough testing, tweaking, and understanding the model’s nuances.

By tailoring transaction monitoring rules and thresholds to specific behaviors and profiles relevant to a firm, AI outpaces manual tuning, reducing the chances of missed risks. This becomes especially crucial for firms navigating new and dynamic spaces, where precise and scalable tools empower smaller teams to implement a robust risk-based approach (RBA), even with limited resources. 

What does this mean for my firm?

Firms looking to deploy an AI-driven transaction monitoring solution should familiarize themselves with the obligations and guidance issued in the regions in which they operate, as these requirements specifically apply to the use of automated systems, bias, and data privacy. They should also ensure they have adequate documentation detailing risk assessments and risk management processes for AI, model governance, model testing and validation, and how the algorithm makes decisions to account for explainability. 

When training AI models, firms should use data from multiple sources, covering all demographics and from across geographies, to avoid bias. Finally, compliance teams should ensure that they have senior management support, carry out due diligence on vendors, and ongoing monitoring and assurance.