In the United States alone, the cost of anti-money laundering (AML) compliance is estimated at $23.5 billion per year and growing, with European banks shouldering a $20 billion burden annually. To counter cost pressure (as well as increasing demands from regulators and customers), many banks have already begun their journey to creating next-generation anti-financial crime programs. They are hoping to boost effectiveness and efficiency.
It’s still early days, however. In general, the industry remains in the disruption phase, characterized by pilots within specific lines of business. These early initiatives have been tremendously useful in learning the art of the possible. They have also helped financial institutions understand and prepare for the “typical” journey to next-generation anti-financial crime.
As someone working with banks along their transformation journey, I aim to share an overview of what a journey typically looks like, and how to ensure success at each phase.
Getting from Pilot to Scale
The journey to next-generation anti-financial crime takes time, planning, and a strong foundation, as leveraging these technologies alone or in siloed initiatives cannot achieve maximum efficiency or effectiveness. Financial institutions need a coordinated approach that can overcome the common obstacles to adopting next-generation technologies at scale:
- Data quality issues: AI and machine learning (ML) can deliver a huge performance boost. However, the accuracy of their results depends on the quality of the models and the data fueling them. Data quality issues are endemic, even in large banks. Many still struggle with unstructured data and the basics of getting a single view of the customer from a risk perspective.
ML initiatives require large volumes of data—both structured and unstructured—for training and running models. Traditional data aggregation via ETL is cumbersome, very expensive, and time-consuming. These realities can present challenges even in the pilot or experimentation phase of an AI initiative.
- Lack of explain-ability: While AI and ML technologies can deliver tremendous benefits, they are of no use if financial institutions cannot explain the models and calculations to regulators. Institutions need to find a way to move beyond these black-box constraints and deliver greater transparency. While regulators are supportive of innovation, institutions must be able to defend the technologies and explain the results to regulators.
- Operationalization: Data scientists lead most AI and ML initiatives. Alerts still require financial investigative unit intervention, however, so coordination among teams is crucial. AI and ML can also create new operational challenges and complexity. These include the user-training challenge of understanding new behaviors, the integration challenge of merging AI/ML detection with case management, and the regulatory challenge of defending the methodology. All of these factors have made operationalization of ML models extremely challenging.
- Inability to scale: A lack of enterprise tools, access to enterprise data, and poor integration of systems have made it extremely difficult for financial institutions to scale AI and ML-driven initiatives beyond a few lines of business.
The Road to a Next-Generation AML Program
Once institutions are equipped with knowledge and insights from the experimentation phase, they require a framework and solutions to operationalize next-generation programs at scale. As their anti-financial crime programs bring in advanced analytics, financial institutions typically move through five phases:
- Initiation - Adoption of rules-based detection and case management: This is the first step of the modernization journey and represents where many smaller financial institutions are today. This approach incorporates the adoption of a rules-based transaction monitoring system and case management capabilities to satisfy standard regulatory requirements. This also helps create a more unified anti-financial crime platform, which lays the foundation for the rest of the journey.
- Incorporating multi-dimensional behavior detection - Typologies-based detection and basic correlation: As organizations look to move up the automation and AI/ML maturity curve, they seek holistic, multi-dimensional monitoring that detects specific typologies (patterns of activity), in contrast to simple rules which monitor individual transactions. This approach enables firms to correlate outputs in a single case and to leverage shared information, such as tax ID numbers, phone numbers, and addresses, to create a more holistic view of each entity. Correlation is critical to enabling institutions to shift from an “alert” to a “case” approach to AML management, achieving mass surveillance - where a comprehensive network of correlated, related entities is monitored rather than individual entities - and reducing the false positive rate.
- Better identification of entities and networks - Enabling graph correlation detection and entity resolution: Once an institution is comfortable with typologies and correlation-based detection and is using a single platform, graph analytics can be incorporated for further effectiveness. Graph analytics is proven to be a game-changer in understanding financial crime behavior. There are two primary areas where financial institutions should consider applying these capabilities in their quest to modernize their AML program.
- Correlation: Instead of basic, rules-based correlation to link various entities, a network can be built by leveraging advanced graph algorithms and graph analytics. For example, a “strongly connected component” graph algorithm can be leveraged to determine strongly related networks and to monitor entire networks. Graph analytics also provides the ability to correlate various entities during detection. Using this technology, firms can further shift from individual entity surveillance toward a mass surveillance approach.
- Entity recognition and resolution: Another use case for graph analytics is entity resolution. Graph matching can provide a holistic view of all matched entities by various attributes, such as name, address, email, and taxpayer ID. The ability to unify data by bringing together entities from multiple internal and external data sources in real time or batch to create a single entity view across the enterprise is vital. Since data quality is fundamental to the success of any ML initiative, institutions that aggregate linked attributes to establish a consistent entity definition will set a strong foundation for the next stages of the journey. Entity resolution tackles the data quality challenge often encountered during the pilot stage and helps build a foundation for quality data.
- Enhanced investigations and optimized outputs - Leveraging supervised ML, contextual investigation, and natural language processing (NLP): After a robust analytical foundation with quality data and graph analytics has been established, additional use cases can be incorporated. Typically, case management and modeling are handled in two separate systems. As a result, deployment of new models and the feedback loop from case management to the analytics systems is an inefficient process. It may even require manual intervention. Continuous discovery and modeling of new criminal behavioral patterns, coupled with the ability to deploy these models rapidly, is a critical requirement in today’s age of continually changing criminal behavior patterns.
Graph analytics and ML are most effective when applied over the data in a comprehensively designed, enterprise-wide financial crimes data lake. This requires a strategic platform designed from the ground up to help data scientists discover and model financial crime patterns faster and more efficiently, thereby increasing their productivity. The platform should support and leverage several types of advanced intelligence capabilities, including:
- Supervised ML: In supervised learning, “the output of your algorithm is already known—just like when a student is learning from an instructor,” explained Bernard Marr, a futurist, author, and expert in digital business transformation in a 2017 Forbes column. “The algorithms are taught from a training data set. If the algorithms produce results that are widely different from what the training data says should be expected, the instructor steps in to guide the algorithm back to the right path.” Guided ML can be applied effectively for case scoring, entity scoring, and correlation scoring. As financial institutions often learn in the pilot phase, however, a lack of explainability will lead to the cancelation of the machine learning initiative, and regulators will not approve it. Therefore, it is fundamental to build high performing white-box models (such as Oracle’s “ModelXray”) to provide insights and transparency into black-box models. This can increase monitoring effectiveness by 40 percent.
- Contextual investigation: Modern solutions should enable the use of powerful graph analytics to connect the dots among customers, accounts, external parties, transactions, and external data. They should also provide a holistic representation of networks that can uncover otherwise hidden suspicious patterns. Investigators can click their way through the entities and their connections, represented as nodes on the graph model, to analyze networks and suspicious activities. These capabilities enable institutions to build networks of activity to gain a greater understanding of customers and their connections, relationships, and behaviors.
- Collective intelligence and collective learning: This capability leverages AI to enhance human expertise through “recommendations” or “next-best actions.” It can be applied to help new or low performing analysts gain situational awareness and learn institutional best practices.
- NLP: This technology can tremendously boost investigator productivity by automatically creating the case narrative required to document and raise a case to the next level of investigation within a firm or to trigger an external regulatory notification.
- Detection of unknown unknowns: Leveraging the infrastructure from previous stages, which creates a strong data provisioning, graph and ML foundation, institutions can continue to enhance their capabilities to discover unknowns to a point never imagined. In this case, firms can detect previously unknown and unmonitored suspicious behaviors leveraging un-supervised ML. Additionally, the capability to detect unknown unknowns can help with assessment of overall inherent risk before the launch of new products and services and allow institutions to plan better controls. Technologies that enable this final leg of the journey include:
- Un-supervised ML: Criminals continue to evolve money-laundering behavior, which means institutions are always trying to catch up with criminal activities, and, depending on the agility of the program, many such activities may go unmonitored. Unsupervised ML can be used to detect these unknown suspicious events.
An unsupervised learning model can churn through vast volumes of data to find an unknown risk that would be nearly impossible to detect using traditional methods. Unsupervised learning does not use a training data set, and outcomes are hidden from the start. According to Marr, “Essentially, the AI goes into the problem blind—with only its faultless logical operations to guide it. … unsupervised machine learning is the ability to solve complex problems using just the input data and the binary on/off logic mechanisms that all computer systems are built. No reference data at all.”
Unsupervised ML models mainly deal with un-labeled data. An institution does not have to know if a given person or entity is good or bad, which ones are “edge cases,” behaving out of the norm relative to their peers.
On the path to maturity, firms should start with semi-supervised models that deliver additional benefits but offer higher levels of control. “Often, … the reference data needed to solve the problem exists, but is in an incomplete or inaccurate state. Semi-supervised learning solutions are deployed here, able to access reference data when it’s available and use unsupervised learning techniques to make ‘best guesses’ when it comes to filling in gaps,” said Marr.
- Sentiment analysis: Sentiment analysis is also known as emotion AI. It can help identify and extract opinions from narrative leveraging the text from blogs, social media, media scans, and more. Firms should look for solutions that provide the ability to incorporate sentiment analysis findings to identify unknown behaviors that can help to drive more informed decisions about potential cases.
Things to Remember as You Begin Your Journey
Financial institutions cannot immediately abandon rules-based monitoring for advanced AI models—but that day is rapidly approaching. To reap the benefits of these powerful new technologies now and in the future, firms must begin the lay the groundwork today. Building a robust foundation is vital for institutions to scale advanced analytics initiatives across the enterprise. Without a strategic roadmap, your program may run into technical limitations. It may not be scale to all lines of business and products. At each phase, however, you’ll want to prioritize flexible technology solutions, purpose-built teams, and collaboration with regulators. Doing so will leave you well-positioned to complete your journey to a next-generation anti-financial crime program.
Message me to learn more or join the conversation by commenting below.
For more information, please visit:
Oracle Financial Crime and Compliance Management Applications: oracle.com/aml
Oracle Financial Services: oracle.com/financial-services
Subscribe to our Blogs:
Oracle Financial Services Blogs: Sign up today