Financial institutions need to move past a rules-based, piecemeal approach to sanctions screening and look to AI for a more holistic and strategic fix.

The punitive consequences of sanctions and anti-money laundering violations for financial institutions (FIs) continue to be significant. Between 2018 and 2020, the total annual penalties issued by regulators globally for such breaches rose from $4.51 billion to $10.4 billion, with one bank alone receiving a record fine of $386 million last year.

Add to this an ever-evolving regulatory landscape and the ongoing financial burden of processing false-positive sanctions alerts, and it becomes clear why sanctions screening remains such a high priority for financial crime teams.

Emerging ESG disclosure responsibilities, along with a heightened focus on responsible banking, will only serve to accentuate the need for FIs to move away from tactical, reactive solutions, toward strategic approaches that can innovatively adapt to further change. In short, it’s no longer sufficient to focus on tweaking existing sanctions systems.

Rules upon rules

What we’ve seen at FIs over the last decade or so is an over-reliance on rules-based approaches to name screening — the process used to identify individuals and companies who appear on sanctions and global law enforcement lists. While any screening platform will inevitably leverage risk-based rules, the challenge comes when these solutions must adapt to changes in lists and regulatory regimes, as the only way this can be handled is typically to introduce additional rules or a whitelist.

This rules-based approach fulfills short-term requirements, but because it is piecemeal and incremental in nature — and results in a proliferation of rules — it creates an overly complex solution, thus increasing reputational and operational risk in the long term. Further, the kind of fine-tuning of applications frequently used to adopt these tactical solutions is finite and can only mask the deficiencies of their underlying approach for so long.

Ethical screening is an added challenge on the horizon that will be exacerbated by the dominance of rules-focused platforms. A good illustration is provided by the challenges faced when screening names from regions with extensive transliteration, such as Slavic and Arabic-speaking countries, where some names can be spelled more than 10 different ways. These difficulties have been exacerbated by the practice of building rules upon rules for name filtering. 

Yes, there are some easy cost-saving wins to be had through the use of robotic process automation to accelerate simple processing and remediation tasks; however, the truly significant gains will be achieved by addressing the more fundamental issues of prevailing name-screening approaches. These include: 

• Only a small amount of the available payment message is extracted by the parsing engines used today.
  
  
• Matching is typically carried out on a one-to-one basis, for example name-to-name or address-to-address, with no possibility of tiered matching of multiple attributes. This can lead to substantial numbers of false alerts, such as when an individual is matched with a legal entity or country. This problem is best illustrated by the often-quoted example of the actor Cuba Gooding Jr., who could be flagged by some systems as a breach of US sanctions against Cuba.
  
  
• When the focus is only on name matching, FIs are unable to harness the wider range of attributes in transaction data available to them through which they could build more sophisticated risk profiles that could be shared and enhanced with datasets from other financial crime domains.

Toward intelligent name screening

To make significant gains in reducing false positives and increasing matching accuracy, FIs must break with the past and reevaluate the foundations on which their screening approach is built. Here are four steps FIs should take:

1. Move away from a rules/whitelist approach. With the continually expanding list of risk entities that FIs must screen for, such as Office of Foreign Assets Control, Politically Exposed Persons and Adverse Media, it’s becoming essential for FIs to use a more sophisticated, multi-tiered approach that harnesses more of the data available to financial crime teams.
   
   A good example is a UK regional bank that asked us to help address the high level of false positives it was experiencing. Working with one of our partner solutions, we helped the bank realize a 70% reduction in these alerts, leveraging a patented phonetic algorithm, historical and enriched datasets and a machine learning-based recommendation engine. This solution applies AI throughout the sanctions value chain, can be easily integrated with the incumbent screening engine and case management components, and provides accuracy uplifts with reduced Level 2 and Level 3 alert processing.
   
   
2. Start building and sharing customer risk profiles. FIs should move away from the one-to-one approach to entity resolution used in rules-based platforms and establish the capability to build out more detailed profiles of customers’ risk, enhanced by external contextual data sources, such as Companies House and Dun & Bradstreet.
   
   When shared within financial crime teams, these risk profiles enable a much more multi-dimensional approach to screening, using a customized Name Entity Recognition model to reclassify entity categorization. When this approach is implemented properly, we’ve seen it address 5% to 10% of false-positive volumes in its own right.
   
   
3. Look beyond industry incumbents to address fundamental problems. It is no longer sufficient for FIs to rely entirely on the established providers of screening platforms to solve long-term industry challenges. These incumbents have demonstrated only incremental evolution, with limited innovation in AI and a focus only on the periphery, such as how to reduce alerts generated by their own system, by creating yet more rules. 
   
   The parsing of payment messages carried out by these mainstream platforms is limited to the fields their systems require, leaving potentially valuable data untouched. If FIs are serious about breaking this vicious circle, reducing whitelists, rules and thereby operational risk, they must address the root cause of false positives and adopt solutions that harness the richness of message data available to them that conveys information like location and historical knowledge gained from previous screening iterations.

Taking a strategic, not tactical, approach

It’s time for FIs to stop simply taking the medicine — tactical improvements to existing platforms using robotics and fine-tuning them — and seek a more holistic and strategic cure. The scope for AI/ML techniques to transform the sanctions landscape has expanded significantly since initial adoption of the mainstream incumbent platforms.

Through a combination of more advanced matching algorithms, the inclusion of temporal profiles to improve pattern recognition models and the increased use of unsupervised machine learning to identify new and evolving risk factors, it is no longer necessary to rely on older rules-based systems. But in order to harness these new capabilities, organizations must be prepared to embrace fundamental change to the way they approach name screening. 

Sanctions screening efforts need to pivot from automated ways of finding new rules for alert suppression, to focusing on the discovery of hidden skeletons while substantially eliminating false alerts. It takes just one existing relationship or transaction to expose FIs to regulatory scrutiny, with those found guilty liable to receive substantial fines, as seen in recent years. Operations and compliance leadership need to address this substantial “fat-tail risk” with forward-looking transformation, instead of piecemeal remediation.