AI can take your AML sanctions and adverse media screening program back to the future

Financial institutions have long been tasked with finding a needle in the haystack when it comes to combatting criminal activities. The programs at work can’t only be acceptable to supervisors and regulators beyond a “tick the box” approach, they must maximize employee productivity, maintain cost efficiencies and ensure a positive customer experience.

The process of screening natural persons, legal entities and transactions applies in a variety of anti-money laundering (AML) contexts. At the foundation there remain two key resource-intensive and expensive processes: sanctions and adverse media screening.

AML regulators and supervisors consider screening an integral part of a satisfactory AML program; however, given its enormous complexity and massive data needs, it’s often marred by mistakes, delays, incorrect decision making, inadequate recordkeeping, and exorbitant costs. As a result, screening typically creates a huge headache to manage properly. The reality is that financial institutions are struggling to bring more efficiency and better processes to their screening systems, understanding that merely throwing human resources at the problem is not the answer. Simply put, there is too much work to do and not enough people to do it.

AI and other innovations to counter financial crime and meet regulatory expectations

The daily life of investigative analysts is often fraught with the challenge of scrutinizing mountains of data from multiple internal and external sources within a short timeframe.

In their roles, analysts must assess risk, eliminate numerous false positives, make quick and correct decisions, and generate detailed reports. The teams overseeing level one sanctions alerts are operating under immense pressure as they struggle to mitigate the risk of missing something that really should be escalated. To complicate matters, when false positives arise, they can severely undermine the effectiveness of risk management and compliance programs. At worst, the delays they engender can lead to an institution being placed on a regulator’s watchlist requiring more diligent supervision.

As artificial intelligence (AI) goes further mainstream, the benefits of employing modern AI-based tools to enhance the sanctions and adverse media screening process are compelling. Imagine having the capacity to “hire” a solution that works around-the-clock, year-round without fatigue or the need for time off. AI can rapidly scale to analyze data, execute research, make decisions on alerts (with a handoff to a human analyst in complex situations), generate and publish detailed reports and leave an audit trail.

This solution also represents a more cost-effective alternative to offshoring, outsourcing and temporary labor.

Altogether, it brings greater speed, accuracy and comprehensiveness, as well as reduced false positives. The application of AI empowers financial institutions to transition from periodic screening to a continuous, real-time monitoring approach. As a result, institutions can swiftly assess relevant new regulatory requirements or, in the event a customer becomes a person of interest or greater risk, allow them to take immediate action to minimize AML risk exposure.

For example, a large North American bank was concerned that its AML team of 10 analysts were being overwhelmed by a daily volume of 600-800 screening alerts to review for names alone. The bank recently implemented an AI solution that resolved such alerts intelligently, dispositioning 95 to 98 percent of them in a highly accurate manner. Moreover, the AI solution automatically provided a text narrative for each alert decision that could be shown to auditors and examiners.

Key hurdles to overcome for AI success

Introducing an AI system into a legacy process can be done relatively quickly and efficiently, and on a reasonable budget, but it must be performed thoughtfully.  A careful approach with modest initial goals often sees success.

“Garbage in, garbage out,” as the old saying goes. The power of AI comes from its ability to train on the financial institution’s core banking data and suspicious activity information to allow for a more accurate and dynamic alert and investigation process. To succeed with any AI system, a huge consideration is comprehensive, quality data that the models rely so heavily on. Data accessibility, sourcing, quality, consistency, privacy, and security are all critical, along with integrating end-to-end workflows to allow for a seamless stream of information.

The process of incorporating AI into a screening system includes several key steps. One is gathering data from various sources. The data should be relevant, diverse, and substantial enough to train a robust AI model. Data cleaning is then needed to remove inaccuracies, duplicates, and inconsistencies.

Also critical is a strong AI model selection process, which should have clearly defined objectives and constraints. It may be useful to start with a simple baseline model to establish a reference point, followed by evaluating different model types to determine which ones perform best on the dataset.

Models must be properly trained and validated, followed by a final evaluation of the test set to assess its performance on unseen data. If an AI system arrives at a decision, it should be explainable, so humans can quickly comprehend the ‘how’ and ‘why’ behind it. This is why relevant performance metrics on both the training and validation sets must be tracked to ensure the model is learning appropriately.

Other important steps include seamless integration of models with existing systems, deployment and testing, user training and adoption, and ensuring continued compliance with all applicable laws and regulations.

Overcoming implementation challenges

AI is not a magic cure-all that can be implemented easily. There are various inherent challenges posed by these tools that must be factored in when employing them. For instance, with regards to the example of the large North American bank, the teams faced the challenge of not running afoul of the institution’s strict information security policies when parsing the data to make it optimized for the AI engine. This required revising the data governance policies and procedures to implement a set of strict privacy controls.

When implementing any AI tool for sanction and adverse media screenings, the most vital principle is to ensure it functions effectively across multiple business lines. An AI tool implemented correctly will seamlessly integrate with other current technology investments.

While advanced AI capabilities hold promise and garner much attention, institutions must strike a balance between hype and reality. Institutions should not overlook the potential benefits of simpler solutions. Assessing the adequacy of existing AI tools before investing in more sophisticated options is prudent.

A financial institution should also test the new technology as much as it needs but ideally avoid long parallel runs in production with their prior system, which adds extra layers of cost and concerns. As with any significant change, senior management’s buy-in is a vital factor for the success of any initiative. This involves the active participation, endorsement, and ongoing support of top-level executives (including resource allocation) — from risk to compliance, HR to privacy — aligning the entire organization toward the final goal.

The benefits of employing AI to enhance the sanctions and adverse media screening processes are evident. While the appeal for automating the function is clear, implementing AI brings its own set of challenges that require a strategic and thoughtful approach. Starting slowly with an ecosystem of partners that bring specific strengths to the table is often the optimal path.

Manish Chopra is Global Head of Risk and Financial Crime Compliance at Capgemini.