A major global bank has a global team of more than 2000 people handling their payment sanction screening operations. The team encountered over 95% false positive rates when screening payments for sanctioned entities and also faced increased operational expenses due to rising payment volumes and growing sanctions lists each year. The bank applied a rule-based approach to the suppression of false positives, but the approach was reactive and did not scale well.
It was crucial for the bank to find a smarter and more proactive way to suppress false positives and manage alerts more efficiently, without raising the level of risk. The solution had to address regulatory concerns and implement explainable false positive reduction models with completed automated decision traceability. It also had to dynamically handle newer false positive patterns, perform much faster than a human operator, and scale to handle increasing hit volume without compromising on performance.
Virtusa automated the sanction screening workflow by leveraging deep-learning-based artificial intelligence techniques and creating a dynamic, scalable, and compliant machine learning solution to detect and handle false positives in sanction screening operations.
The bank needed to scale out to address the increasing payment volumes and manage false positives more efficiently.
The key challenges:
- The operational expenses for sanction screening operations was high, which included managing a global team of more than 2500 people dedicated to managing alerts.
- The false positive rate in sanction screening was over 95%.
- The payment volumes were increasing at the rate of 7%+ year-on-year (YoY) and sanction list size was increasing at 10%+ YoY.
- The legacy sanction filtering system was based on static-rule-based filtering algorithms.
Virtusa analyzed the bank’s existing sanction screening operations and proposed a scalable AI solution that helped the bank to reduce cost as well as achieve stronger sanction screening accuracy.
- An automated level 1 review workflow of sanction screening operations by leveraging sophisticated deep-learning-based AI models.
- Using neural networks, Virtusa developed Named Entity Recognition (NER) models to accurately identify named entities in payment messages and sanction lists, and mimic operator decisions.
- To validate the output of each other and ensure high quality of false positive recommendations, we leveraged multiple machine learning models.
- To ensure seamless regulatory approvals, we built in-model explain-ability and decision traceability.
- The solution was built on open source components and ensured on-prem infrastructure deployments.
With intelligent automation, the bank no longer has to worry about increasing false positives or sanction screening accuracy. The solution helped them boost the speed of sanction screening and gain productivity improvement.
The outcome was:
- A cost reduction of over 80% of the L1 manual effort.
- More than 98% accuracy rates.
- Over 2x faster sanction screening process cycle.
- More than 50% additional operator bandwidth with AML investigations for true positive cases.