<img src="http://www.explore-123.com/9942.png" alt="" style="display:none;">

How AI can address the increasing complexity of false positives in  sanctions screening

By Parth Desai

Financial crime compliance across the world is continuing to grow in complexity. All players involved in international business – banks, payment service providers or corporate organisations – must stay abreast of a constantly changing landscape. The increasing political use of highly targeted sanctions restrictions, for example with new US sanctions guidance issued in 2020, combined with ever more resourceful and sophisticated bad actors and terrorists, pose challenges for all organisations. Any failure to comply with sanctions and money laundering obligations brings the risk of severe financial and reputational damage.

The financial cost of non-compliance can be high. Fines issued by regulatory bodies including OFAC (the US Department of the Treasury’s Office of Foreign Assets Control) for failure to comply in the past decade exceeded $26 billion. While the total value of OFAC fines in 2020 was down dramatically on the previous year at $23.6m versus nearly $1.3bn in 2019, it doesn’t mean enforcement is being relaxed. What made the total fines issued in 2019 so large was the scale of the fines issued to two very large European financial institutions. While banks avoided such large fines in 2020, there is no scope to relax vigilance in the year ahead.


Addressing False Positives

It is no wonder that financial institutions dedicate a huge amount of both human and financial resources to ensuring they remain fully compliant, with compliance costs increasing by 70% since the financial crisis. A challenge many face, however, is that the limitations of traditional compliance techniques in the face of growing complexity, has led to a substantial rise in false positive hits whose cost is estimated at USD 2.6 billion annually. Of the alerts generated, often less than 1% represent real financial crime cases, meaning that banks have to spend 80% of their financial crime compliance resources time manually reviewingand ruling out 99% of hits. This places a considerable operational burden on all financial institutions as over 90% of the time of investigators is spent on false positive alerts. Many of the current systems in the market today use a traditional and static rules and token based approach– and are limited in their capability to assist compliance professionals in processing and checking rising false positives in a constantly changing world.

Today there is a growing appreciation of  the value Artificial Intelligence disciplines can bring in ensuring compliance, combatting financial crime, and lowering operational costs. In particular, many organisations have discovered first-hand the significant benefits of AI systems in increasing the accuracy of detection, drastically reducing false alerts and cutting overall compliance costs.

Natural Language Processing (NLP) and Machine Learning (ML) have a key role to play in increasing the accuracy of financial crime detection, and the use of machine learning capabilities can significantly reduce the manual effort required in reviewing false alerts. This intelligent AI-driven dual approach can drastically cut compliance costs while delivering reputational protection across all payment processes and counterparties, with a significant reduction in manual effort.


An intelligent approach, powered by AI

Combining the benefits of NLP, Knowledge Based Systems, and Machine Learning, is very powerful and can provide full explanations for alert review and internal and regulatory audits.

Natural Language Processing outlines the algorithms that mimic human free format language understanding, incorporating context and common sense while processing payments. Using the context surrounding the hits, these algorithms can accurately decipher if a hit is true or false and can explain its reasoning.

Knowledge Based Systems represent a library of rules and information built using years of experience in payments and compliance domains, for example a dictionary of banks, cities, countries and common-sense phrases with meanings which help us understand free format information that establishes context and relationships between them in the same way that a human would.

Machine Learning is the ability of the system to analyse, understand and learn from historical information and past human actions and continuously strive to improve the results and improve the detection of False Positive hits.

This AI based approach has reduced over 75% of false positives, has reduced the time to process each alert by 80%,  has increased productivity by 400% and last but not the least, is 100% accurate.


Explainable and Auditable

Any AI-based solution should be explainable and auditable as financial institutions need to understand and be confident about the actions being taken by the AI-based solutions and to explain to the auditors and regulators the reasons for all decisions taken.

Pelican Secure Self Learning Optimisation applies exactly this approach to sanctions screening . It combines the power of NLP, Machine Learning and Knowledge Based Systems, not only to understand and classify false positive alerts generated by third-party tools, but also to explain each decision made by the system. This allows compliance staff to resolve false positives much more quickly – reducing inefficiencies and freeing up valuable resources.

The full audit trail provides detailed evidence for regulators or any future investigations. Pelican has successfully demonstrated the powerful capabilities of its technology, with Pelican Secure Self  Learning Optimisation being able to classify over 75% of the false positives for a large financial institution. Pelican Secure Self Learning Optimisation is a mature and industry-proven solution that works with a range of existing third-party financial crime compliance tools. The underlying Pelican AI technology platform which uses Machine Learning and NLP, has been live across a number of global banks for over 25 years, processing over one billion transactions worth over US$5 trillion.

The message is clear:  an AI-powered approach, can help banks overcome some of the most complex financial crime compliance challenges they face today and to do so in a way that frees up valuable resource to focus on investigating any true positives that do arise.