Anatomy of a modern AML program
1. A consolidated back end
The first step in the journey to AML program modernization is to consolidate the back end and move from disparate systems to a unified platform for KYC/customer due diligence (CDD), monitoring, detection, investigation, and reporting. A unified platform provides several benefits. First, it improves decision accuracy because it allows AML investigators and analysts to evaluate the risk of each event holistically, regardless of where it comes from. Second, a standardized, unified platform helps financial institutions control costs by providing operational efficiencies, reducing the training costs required to keep staff up-to-date on multiple systems, and making daily workflows smoother. Finally, a unified platform can give CCOs a view of end-to-end compliance operations, which they can measure and manage within a single dashboard.
2. Unified and context-appropriate data
A modern AML program requires more than a unified platform; it also needs unified and context-appropriate data. This can be achieved by having a common data foundation that can take input from any transaction system and any data source, including third-party data feeds and fragmented data. A common data foundation serves as a single source of truth that facilitates quality control and enables consistency, transparency, and auditability. To truly modernize the AML process, data must be context-appropriate. It’s not enough to simply aggregate information on a customer origination. To truly bring new levels of effectiveness to the AML process, financial institutions must also match sentiment to that data, extending well beyond fuzzy matching. With unified and context-appropriate data, financial institutions can source data once and use it for multiple use cases, instead of performing extensive ETL cycles to make data available in the right location at the right time. Importantly, unified data makes it easy to leverage the detection data pipeline for discovery and modeling of new criminal patterns, and for advanced analytics applications that enhance monitoring, detection, and investigation results.
3. Advanced analytics
While many financial institutions are excited about the potential of advanced analytics to dramatically improve the effectiveness and efficiency of their AML programs, they often wonder where to start. Fortunately, advanced analytics can be brought in gradually. Each financial institution can take the path and pace that works for them based on their needs and their data pipeline.
Advanced analytics to consider as part of an AML program modernization initiative include:
- Machine learning to improve detection
While traditional rules-based AML scenarios may keep financial institutions technically compliant, they are unable to adapt to the constantly changing patterns of modern financial crimes, elevating risk to the business and its reputation. Machine-learning models can improve detection by rapidly adapting to changing patterns.
To start, financial institutions can run models in parallel with rules-based scenarios and eventually turn off the rules when their regulators are comfortable. Machine learning models can be overlaid on top of rules-based scenarios to detect highly suspicious activity. They can also simplify tuning by replacing multiple rules and eliminating the need to tune multiple parameters in favor of a single probability score.
- Graph analytics for better investigations
Efficient investigation of highly organized financial crime requires technologies such as graph analytics (PDF) to succinctly express intricate money movement patterns, detect multi-hop relationships, and identify hubs and spokes of activity. Graph analytics leveraging a single source of data powers investigators with an ability to search customer information from various source systems and allows the linkage of customers, accounts, external entities, transactions, and external data stored in disparate operational silos. A single source also provides a 360-degree view of a customer, foreign body, or account for a holistic view of the case, transactions, and external data of interest. Graph algorithms, such as connected components and shortest path, can generate automatic linkages. For example, such linkages could be based on customer identification numbers, name-matching, shared phone numbers, tax IDs, and more.
Furthermore, investigators can drill down (expand/collapse) on customer information and visualize related parties using graph analytics. Graph analytics can be combined with machine learning to build better cases by correlating red flags and suspicious events from various systems into a single case. This allows investigators to work on comprehensive cases instead of single events, which significantly reduces AML program workload.
- Entity resolution for a 360-degree view
Graph analytics enables entity resolution, which allows institutions to gain a genuine 360-degree picture of their customers and external entities, alike, by identifying different instances of the same entity across data sources.
- Deep learning to find patterns
Deep learning in financial crime can be valuable because similar past behaviors can be clusters that can predict new cases. Financial institutions can apply deep learning to graphs to find new graphs that are similar to previously identified graphs of criminal activity. Deep learning on graphs is especially effective because it is not limited to a specific point in time for risk indicators but instead evaluates holistic patterns and networks.
- NLP for automatic case narratives
Financial institutions can use natural language processing (NLP) to make graph-based investigations more efficient. NLP can automatically generate case narratives based on what an investigator uncovers in a graph visualization. This eliminates the manual step of writing case narratives, thereby reducing investigation times dramatically and avoiding human errors.
- Collective intelligence and collective learning for recommendations
Financial institutions can use artificial intelligence to learn from previous case decisions about graph networks, and provide recommendations or suggest next steps to investigators. This can help new analysts or investigators learn to identify criminal behavior more quickly.