Khurram Akhtar — Cofounder of ProgrammersForce.
How smart is your AI-powered tool if the data it relies on is flawed?
AI, and especially agentic AI tools, are evolving at a rapid speed, with the use of AI increasing from 33% in 2023 to 71% as of July 2024, transforming industries with their predictive power and automation. But beneath the surface, their true effectiveness hinges on one critical factor: the quality and structure of the data used to train them.
Businesses using AI-powered tools often face a frustrating paradox: Despite using advanced algorithms, they may still receive inaccurate or unrepresentative results. This issue becomes even more paramount for businesses using anti-money laundering (AML) tools, where the backbone of the AI solution, an AML data layer, is often weak or poorly structured.
I have noticed that a key challenge businesses (obligated to comply with AML laws) face is that legacy AML compliance regtech solutions are built on outdated, poorly structured datasets that fail to account for jurisdictional nuances, evolving regulations and contextual risk factors, resulting in higher false positives.
Moreover, I believe the lack of harmonization and context-awareness in data layering is a common, but critical, flaw in many AML compliance tools. Let’s examine how inaccuracies in AI-powered tools can hinder businesses from fulfilling their compliance and risk management commitments.
Issues In Data Layering For AI Tools: The Risk Of Misclassification Trap
Inefficient data layering leads to a failure in accurately categorizing politically exposed persons (PEPs) levels, which directly impacts risk scoring. Many tools do not distinguish between high-risk PEPs, like congress or senate members, and low-risk ones, like municipal mayors, especially across different jurisdictions.
This lack of structured PEP stratification means banks and fintechs can receive skewed risk scores, making it harder to prioritize due diligence and ensure compliance in line with jurisdiction-specific laws without unnecessary friction or exposure.
Jurisdictional role ambiguity adds another layer of complexity. Political roles vary across countries, yet many tools use a uniform scoring model. A “regional council member” in one country might wield substantial influence, while the same title elsewhere is largely ceremonial.
This can make it difficult for financial institutions and obligated sectors operating worldwide to comply with global AML regulations, as the tool does not account for diverse data from different jurisdictions.
Sanctions Compliance Solutions: The Lag In Risk Assessments
Many compliance systems rely on tools that simply offer the official sanction lists, like those offered by OFAC, and do not incorporate adverse media screening when assessing individuals. Important updates, such as press releases naming high-risk individuals, are often missed or incorporated into watchlists weeks later.
Additionally, accurate and up-to-date sanctions data is also important for financial institutions, especially instant payment service providers in the European Union, for whom daily and speedy screening against EU restrictive measures is now a legal obligation.
AI Automation And Updation: Scoring On Shaky Grounds
AI-powered compliance tools are only as good as the data they’re fed. When data is poorly structured or mislabeled, risk scoring becomes flawed and unreliable. Say an AI model misclassifies the risks associated with a politically exposed banker due to ambiguous metadata, the system may either clear them or flag them inaccurately, leading to higher false positives.
Jurisdictional Data: When The Map Misleads The Model
Legacy systems often lack nuanced jurisdictional data, especially for smaller or conflict-affected countries. This results in blanket assumptions or missing entries.
For example, a financial entity operating within the USA may fail to identify a PEP from a war-torn country simply because the jurisdiction is unrepresented in the dataset. This often leaves a serious risk undetected due to poor geographic tagging and coverage.
How Leaders Can Ensure Accurate AML Risk Scores
1. Demand custom scoring models with flexible architecture.
Default risk scoring models often fail to reflect flexibility according to your business’s unique risk appetite. Choose tools that allow custom scoring rules, adapting to specific sectors, regions and risk thresholds. Additionally, opt for tools that offer customized user journeys. For instance, a fintech company operating in both Canada and the U.K. must account for different PEP classifications.
2. Prioritize tools with proactive adverse media screening.
Ensure your compliance system doesn’t wait for individuals to be added to a list before flagging them. Opt for solutions that utilize real-time adverse media scanning from reputable sources (including press releases and government reports) to catch high-risk individuals early in the risk cycle. This ensures that you do not have to indulge in risky transactions and dealings with clients who are involved in any malicious activity.
3. Audit your AI inputs regularly.
AI automation isn’t a magic bullet; it needs accurate, well-labeled data to function correctly. Conduct periodic audits to check for misclassified roles, ambiguous metadata and gaps in labeling that could skew your risk outputs. An excellent AML tool should ideally allow testing. Always compare multiple vendors before you finalize on your option.
4. Invest in dynamic, context-aware data layering.
Static PEP and sanctions lists may no longer be sufficient. Business leaders can seek AML tools that use real-time data updates, jurisdiction-sensitive categorization and life cycle tracking for PEPs. This can ensure customers are properly identified and assessed based on current, relevant risk levels.
I advise all businesses to always ask their vendor: How relevant is the data of your AI-powered compliance solutions, and do you distinguish between PEP levels and jurisdictions?
Accurate AML Starts With Data—And Better Structuring
In compliance, AI matters—but accurate data matters more. To ensure AML compliance that is in line with the Financial Action Task Force and FinCEN requirements for a risk-based approach, the use of technology that accurately assesses the risk associated with an individual is required.
I often say to my fellow businessmen: Accurate AML starts with data, but it’s the structuring that determines its impact. Poorly layered data leads to three things: more false positives, greater manual review time and lower return on investments. And no one wants that!
Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?
Read the full article here
