The financial services industry is undergoing a new transformation driven by artificial intelligence (AI) technologies, especially machine learning (ML). Compliance solutions powered by these technologies for anti-money laundering (AML), counter-financing of terrorism (CFT), sanctions screening, and so on as part of customer lifecycle management (CLM) are rapidly enhancing risk detection and mitigation capabilities.
We wanted to find out more about the possibilities and practical applications for these AI-driven solutions. So, we asked a data scientist in the banking sector six key questions about AI’s role in financial crime detection and prevention.
Nuray Yücesoy, MSc, Big Data Analytics and Management at BNP Paribas, talked to Moody’s Industry Practice Lead, Francis Marinier, and Senior Solutions Specialist, Nicolas Pintart, about the different layers of AI, her experiences of importing AI into AML control environments, and her view of the challenges and opportunities for early adopters.
Nuray Yücesoy: The conservative nature of compliance groups within the banking sector can make adoption of new technologies challenging. Fear of new practices resulting from AI and a lack of detailed guidance from regulators pose hurdles. Despite regulators encouraging new technologies, like AI, their stance isn’t fully detailed. Uncertainty in this area poses risks for early adopters who are eager to progress but who may experience uncertain outcomes and approvals related to their innovations.
There is a growing need for compliance officers who are versed in regulatory frameworks and who are also innovative, data-driven, and capable of navigating the nuanced challenges of modern finance. Their expertise will be crucial in fostering an environment where compliance is seen not just as a regulatory requirement but a dynamic component of strategic decision-making and a way to grow the business.
Francis Marinier: AI in financial crime compliance operates at various levels, from data preparation and cleaning to detection and investigation. Traditional methods often use AI as an additional layer on top of rule-based systems, but a more holistic approach is emerging.
Nicolas Pintart: The new approach to anti-financial crime processes starts with AI and ML instead of them being leveraged as an afterthought. It is more efficient this way in detecting new patterns, minimizing false positives, and optimizing operational efficiency. The key is to have a unified view of compliance and an open mind to creating the most efficient and effective approach. This emphasizes the importance of collaboration between solution providers, financial institutions, and regulators.
Thanks to early adopters in financial services, vendors have been able to enhance their AI/ML tools. In return, financial institutions have had the opportunity to customize new tools according to their specific needs, essentially shaping the future of compliance with their vision.
Francis Marinier: The success of AI/ML projects hinges on data quality and completeness, which affects the accuracy, reliability, and effectiveness of models. Improving data processes and integrating external data sources are crucial for enhancing the quality and effectiveness of any AI/ML model for today and the future.
Nicolas Pintart: Ethical concerns and bias in AI models are significant issues that can undermine the fairness, transparency, and accountability of AI systems in financial crime compliance.
Nuray Yücesoy: Many people have concerns over use of AI leading to a lack of transparency around decision making and audit. It is possible to achieve transparency in AI-driven compliance processing and transaction monitoring.
Explainability involves understanding how algorithms work, their purpose, and the rationale behind their outputs alongside a proper log of events for post-process verification. This clarity is essential for regulatory compliance and ethical considerations. The ACPR Governance of Artificial Intelligence in Finance June 2020, provides the requirements of the regulator on this subject.
Nuray Yücesoy: First, strategic data management is crucial. Effective AI-driven compliance begins with a robust data architecture. Organizations must invest in scalable and flexible infrastructures capable of managing the volume, velocity, and variety of financial data while establishing strong governance practices to ensure data quality and integrity, which are essential for training reliable AI models.
Secondly, AI should not operate in isolation. An integrated control environment that aligns AI initiatives with broader regulatory and compliance frameworks is vital. This involves developing AI tools and models that support compliance objectives and which are transparent and explainable to regulators. Providers and compliance officers need to be aware of the existence of various AI-led models, along with their relative strengths and weaknesses.
Finally, AI models in financial crime solutions must continuously evolve to respond to emerging threats and changing regulation. This requires ongoing training, fine-tuning, and validation to maintain relevance and effectiveness.
These processes should be reinforced with internal procedures and audit. The engagement with AI in anti-financial crime solutions offers significant opportunities for enhancing efficiency and effectiveness. However, the benefits can only be realized through careful planning, integration, and management of both the technological and human elements of AI initiatives.
By focusing on these areas, compliance professionals can harness the potential of AI to create more robust and responsive compliance environments that bring their experienced and expert teams on the journey.
Nicolas Pintart: Bias in AI can originate from various sources, including but not limited to:
Training data, if not carefully sourced and curated, can contain historical and statistical bias that AI models can inadvertently learn and perpetuate. Data collection is critical to the success of the models.
Nuray Yücesoy: Bias often stems from the data used to train AI models, reflecting historical inequalities or present-day disparities. The impact of these biases can be profound, leading to discriminatory outcomes or unfair treatment of certain groups, potentially exacerbating social inequities.
However, we can tackle these issues and design solutions that don’t rely on information such as gender, age, or nationality for example. For an effective AML-CFT solution, these details are not mandatory. By proactively addressing the issue of bias, financial institutions can harness the power of AI in compliance fairly and with inclusivity.
Adopting transparent AI methodologies that allow for the examination and understanding of how alerts are generated can facilitate trust among users and stakeholders. Feature engineering has a critical role at this stage. From my point of view, supervised models that are too dependent on historical decisions could carry a higher risk of bias compared to semi-supervised models and unsupervised models. This is due to the fact that historical data carry biases which could undermine AI-based AML systems.
Nuray Yücesoy: AI, especially with ML technology, is redefining financial crime compliance, offering sophisticated solutions to prevent and detect crimes faster, more effectively, and more consistently.
Despite the challenges, the potential benefits from AI for both regulatory compliance and societal well-being are immense. As the industry navigates these technologies, three factors become key to all data scientists working on AI development:
Nuray Yücesoy: The core of effective AI-driven compliance solutions is high-quality data. I advocate for data solutions that provide comprehensive coverage, high granularity, and frequent updates, including contextual information that can significantly enhance decision-making processes.
Furthermore, the integration capabilities of these solutions is paramount. They should seamlessly integrate with existing systems through APIs and modular solutions tailored to meet the specific needs of diverse technical infrastructures.
Given the increased scrutiny of AI systems by regulators, I also seek solutions that prioritize transparency and explainability, with clear documentation of data sources and methodologies that are comprehensible not just to data scientists but to non-technical stakeholders as well.
The dynamic nature of financial crimes and changing regulatory landscapes necessitate continuous support and collaboration from data providers. We need partners who are committed to providing ongoing support, training, and consultation to help navigate new challenges and adapt to regulatory and technical changes.
By addressing these needs, data and analytics solution providers can equip compliance professionals with the necessary tools to effectively navigate today’s complex environments and prepare us to meet the challenges of tomorrow.
Our discussion highlighted the ongoing and certain integration of AI technologies into financial crime risk management and compliance. There needs to be a focus on the balance between enhancing risk detection and addressing challenges such as data quality, ethical considerations, and the changing needs of regulatory compliance.
Key to successful adoption and use are a strategic approach to AI, one that emphasizes the mitigation of bias, ensures transparency, and provides an exhaustive audit trail while being adaptable to changes. Also, collaboration among stakeholders is needed to leverage AI's full potential in combating financial crimes.
The future of AI in compliance has clear promise, but requires continuous innovation, ethical vigilance, and collective effort to navigate its complexities effectively.
Thank you to all the contributors, and if this paper raises any questions for you, please get in touch with the Moody’s compliance and third-party risk management team.
Moody’s is transforming compliance and third-party risk management. Integrating award-winning data, workflow automation, and AI-driven solutions, we are creating a world where risk is understood so decisions can be made with confidence.
With innovative technology and industry expertise, Moody’s automates perpetual monitoring of counterparty risk across global networks in near real-time. We work with customers to shape their know your customer (KYC), anti-financial crime, risk, and compliance programs around their risk appetite, operational needs, and strategic goals.
Customers build a unique risk management ecosystem, tailored around their policies and compliance obligations. Leveraging Moody’s digital-first solutions for efficiency, scalability, and flexibility, customers manage processes from customer and supplier onboarding to enhanced due diligence and risk monitoring.
Moody’s is helping customers build a picture of risk across 197 countries, and 211 jurisdictions, screening against our database of +21 million risk profiles, +489 million entities, and +51,000 sanctioned entities.
Our data-driven solutions empower risk and compliance professionals to make decisions efficiently and effectively, using a risk-based approach. Staying ahead of risks and ensuring the integrity of operations.