Using AI to reduce false positives in anti-money laundering

Global interconnectedness made possible by digital transformation is driving growth opportunities for organizations across industries. In banking, it’s leading to borderless transactions, digital currencies, and real-time payments, as well as more stringent regulatory scrutiny. These changes present new risks, putting pressure on banks to evolve their anti-money laundering (AML) programs to meet rising compliance demands.

Traditional screening methods, built for simpler, slower systems, can struggle to keep pace with today’s complex, high-speed digital environment. This mismatch can lead to elevated rates of false positives that disrupt the customer experience, inflate operational costs, increase regulatory scrutiny, and compromise business growth as banks balance compliance demands with operational stability.

This situation also raises important questions:

• How can your organization protect customers from the impact of false positives?
• Can AI effectively reduce false positives and improve compliance?
• How can AI be configured to accommodate complying with ever-changing regulations effectively?

AI is becoming increasingly valuable in helping banks combat financial crime and comply with the complexities of evolving regulatory standards. More specifically, through machine learning and other advanced capabilities, AI can reduce false positives, helping banks save costs while strengthening their compliance posture and outcomes.

Hidden costs of false positives in sanctions screening

In today’s digital landscape, a false positive occurs when a legitimate transaction or customer is mistakenly flagged as involved in sanctioned or illegal activities. This misidentification, often termed a “false alarm,” results in the denial of services or transactions to innocent customers, causing inconvenience and frustration.

False positives can be as damaging as false negatives—instances where genuinely suspicious transactions are mistakenly approved. In the financial sector, this issue is particularly pronounced; studies reveal that false positive rates in banking can reach as high as 90%, leading to frequent transaction resubmissions and damaging brand reputation.

This challenge isn’t limited to traditional banks. Digital banks, credit unions, payment service providers, and fintech companies alike are affected, making it a widespread concern across the financial industry.

Why false positives can be more costly than violations

False accusations or wrongful transaction denials can severely undermine customer trust and loyalty, often prompting customers to seek alternative solutions with competitors. The financial impact of false positives includes increased customer service inquiries, transaction resubmissions, and potential legal liabilities—all of which can significantly erode profitability.

The regulatory scrutiny and long-term reputational damage associated with false positives can be even more severe than the consequences of a sanction violation. The added burden of data privacy obligations, such as GDPR, further complicates matters: improper handling or misidentification of customer data may lead to privacy breaches and fines. Regulatory audits, obligations to strengthen internal controls, and increased reporting requirements can compound these issues, creating cumulative costs that often surpass the penalties or reputational harm from an actual sanction breach.

This makes it essential to prioritize the reduction of false positives to maintain customer trust, protect reputation, and minimize regulatory risk.

AI as a viable solution to minimizing false positives

AI supports intelligent adaptability, empowering an AML detection system to dynamically adjust to the unique demands of its operational environment and evolving AML risks. With AI-powered flexibility, the system can continuously calibrate in response to shifting data patterns, delivering responsiveness and precision in detecting sanctions and minimizing false positives.

These outcomes can be best achieved through an integrated framework that combines advanced AI techniques, including reinforcement learning, Markov decision processes, and reward engineering.

1. Reinforcement learning: This technique enables an AML detection system to continuously learn from its environment. By interacting with data, the system’s AI model receives feedback in the form of rewards or penalties based on its decisions. Over time, this learning process enables the model to refine its strategies, improve its decision-making, and better identify true positives.
2. Markov decision processes: These processes offer a structured approach to modeling the decision-making process involved in sanctions screening. By defining key elements such as states, actions, transitions, and rewards, this approach enables a systematic evaluation of strategies and potential outcomes, enhancing the AI model’s ability to make well-informed and adaptive decisions.
3. Reward engineering: This involves designing a reward function to guide the system’s learning journey. By carefully structuring rewards, accurate classifications can be prioritized, minimizing false positives while preserving true positives. Incremental feedback and shaped rewards create a feedback loop that continuously optimizes decision-making to improve accuracy even in complex scenarios.

The table below illustrates how reinforcement learning combined with Markov decision processes and reward engineering enables an AML detection system to learn and refine its decision-making capability.

Combating AML through AI innovation and partnership

Staying ahead of evolving compliance demands requires innovation and expertise. AI delivers the necessary innovation while a strategic AI partner provides the expertise and solutions.

CGI works with banks to integrate AI into their AML detection operations. CGI Hotscan360, for example, equips banks with adaptive, AI-driven intelligence, and it’s backed by the consulting expertise and tailored insights of our AML and fraud detection specialists. To learn more about the power of AI to transform fraud and AML detection and our work in this area, feel free to contact me.