Adjust Your Mindset to Better Handle These New Challenges
Compliance organizations are rightfully concerned about the required growth of their analyst teams and IT infrastructure to handle the increasing regulatory burden. The use of strictly traditional systems and tools would likely lead down that path, raising the “cost of doing business” potentially to values high enough to justify exiting certain markets or jurisdictions. But the past need not be prologue, and savvy managers are becoming aware of fundamental truths that take advantage of recent technological innovations in artificial intelligence and distributed computing:
1. Next-generation know your customer (KYC) and AML operations do not focus on discrete and siloed client risks (e.g., transaction monitoring and sanctions screening), but rather consider a dynamic risk-based view of each client and their context within the broader ecosystem.
2. There is no tradeoff or incompatibility between artificial intelligence / machine learning (AI/ML) techniques and regulator-ready auditability and evidence packages.
3. Advanced system architectures exist to bring together as much data as possible to train algorithms while not compromising on security practices, exposing private information, or breaching local jurisdictional rules.
A Holistic View of a Client
Financial institutions have built compliance divisions over time that adhere to outdated software constraints. This has led to fragmentation of client data across different teams, business processes, and systems, such as KYC, CDD, EDD, sanction screening, negative news screening, transaction monitoring, and others, which make it difficult or even impossible to generate a true, unified view of a client’s risk profile.
Rather, these systems should generate the true business and compliance need: a dynamic view of a client based on all available internal and external data. Combine data from transactions, sources and uses of funds, sanction lists, news articles across multiple languages, and CRM information. Keep the data image current with near real time updates. Maintain the network graph of all related parties, associations, and money flows. Adding additional intelligence layers through advanced rules or AI/ML will then yield tremendous business operational benefits.
Artificial Intelligence is Now Transparent
Financial crimes regulators have high reporting standards and place a significant burden on financial institutions to provide clear, transparent and traceable rationale for suspicious activity reports. These reporting standards have led financial institutions to develop rigid, rules-based systems to detect potentially suspicious behaviors. These systems rely on experts to define customer segments, scenarios or rules and thresholds. When thresholds are exceeded for the rules applied to a particular client, an alert is created. While rules-based systems may enable AML investigators to identify some illicit activity, they often have critical issues.
Furthermore, with rules-based systems, compliance and AML teams must manage an unsustainable number of false positive alerts. The rigidity of the rules does not allow these systems sufficient flexibility to learn behavioral patterns and accurately differentiate normal behavior for one client or segment from suspicious activity. For every productive or necessary investigation, investigation teams may disposition hundreds of superfluous alerts.
Yet these rules-based systems continue to be used despite the issues they create for AML teams. The continued use of rules-based systems is primarily due to the fact that rules are easy to tune, document and explain to regulators, and alerts produced by the system are straightforward and traceable to specific source data.
In the past few years enterprise artificial intelligence (AI) systems have made considerable progress in providing interpretability that will satisfy regulatory scrutiny. At this point, there should be no discernable difference in understanding “how” or “why” an alert was triggered between a rulesbased system or one powered by advanced enterprise AI. As AI systems continue to make progress in interpretability, traceability, data lineage and model management, adoption of AI-based AML modeling techniques is increasing. Additionally, AI-platforms that leverage these techniques can also provide advanced model management and monitoring capabilities that automatically track model performance and store and version key information about models, predictions, risk drivers, alerts and source data. These capabilities support all regulatory and audit-driven requirements for explainability and transparency.
Scale Out Your AI Systems Without Requiring New IT Projects Each Time
Traditional AML systems have often been implemented as standalone technology stacks for each line of business, in every region and regulatory jurisdiction. With this approach there are limited economies of scale for data processing and cross-jurisdictional learnings related to emerging money laundering schemes, or the identification of nested account relationships and global financial crime networks.
However, new advances in distributed computing make global deployments easier, and support combining key support unifying global, cross-border data across jurisdictions and implementations to uncover complex schemes and support more robust AI/ML model training. Some financial institutions are centrally managing their AI models, deploying artifacts locally and enabling regional teams to retrain models based on local data.
Such techniques are enabled by advanced system achitecutres that support global deployments in a much more cost effective way than individual, standalone IT implementations.