Case study

A leap in sanctions screening accuracy and efficiency

How a global investment management company harnessed generative AI to fast-track sanctions alert reviews

Who we worked with

A global investment management company.

What the company needed

To handle a high volume of false-positive sanction screening alerts in its financial crime risk management
To focus analysts' attention on legitimate potential sanctions concerns and violations, not false positives
A cost-effective and efficient software solution for financial crime risk management (FCRM) tasks

How we helped

Architected a technology solution that integrates our financial technology suite, riskCanvas™, with generative AI (gen AI) capabilities from AWS Bedrock
Implemented a system where gen AI provides initial decision summaries based on existing reference data, which the analysts then review and tailor for accuracy and completeness
Fine-tuned large language models (LLMs) with in-house data for effective prompt engineering, improving the accuracy of reports

What the company got

A minimum viable product (MVP) in just 36 hours, with the potential for further efficiencies using straight-through processing (STP)
A significant reduction in the average handle time (AHT), with up to 80% less time spent on reviewing potential screening alerts and writing justifications
Better investigation efficiency, robust reporting, pattern and trend identification, and the ability to navigate changing threat landscapes
$200,000–$300,000 in operational savings
50% reduction in total cost of ownership for screening analysts

The challenge

A manual, time-consuming approach to screening sanctions alerts

Sanctions screening – a process financial institutions use to identify and block transactions involving individuals or entities on international sanctions lists – is an essential part of fighting financial crime today. However, data complexity and volume and an increase in false positives demand innovative strategies to manage sanctions screening efficiently and effectively.

At the global investment management company, analysts were manually reviewing hundreds to thousands of sanctions screening alerts every day to identify bad actors.

Unfortunately, these alerts had a high false-positive rate. Analysts were wasting time evaluating false positives and generating detailed, data-rich, labor-intensive reports explaining their decisions. This distracted from the more careful evaluation of actual sanctions concerns and violations.

Frustrated by these inefficiencies, the global investment management company wanted to optimize the sanctions screening process to save analysts' time and improve accuracy.

The solution

A gen AI-led approach to improve decisions

Genpact was already providing the company with a suite of related services, including transaction monitoring, case management, client screening, know your customer (KYC), and risk scoring. The client banked on Genpact's reputation and financial industry expertise to develop a solution that would address its challenges in managing sanctions screening alerts.

As a first step toward optimizing the report-writing process, Genpact assessed that documentation tasks were a good fit for content-generating generative AI technology. Building on its relationship with AWS, Genpact leveraged its internal robust industry expertise in financial services to integrate Amazon Bedrock with riskCanvas.

In a joint working session with the investment management company and AWS, Genpact refined its prompt-engineering protocols using proprietary fields from the investment management company. We conducted a comprehensive technical review of AWS Bedrock to minimize security gaps and ensure that data in transit is encrypted end to end and remains within the AWS backbone network.

User acceptance testing (UAT) helped determine the optimal dataset for prompt engineering. Additionally, Genpact used a human-in-the-loop approach to ensure that no decisions are made or records updated until the analyst approves or edits the generated decision and summary.

The impact

A more efficient use of analyst resources

The generative AI-led solution is helping the team write summaries of match results that automatically include a suggested decision for sanctions screening alerts. An analyst acts as a human in the loop, reviewing and closing the updated AI-generated narratives.

This is helping the company enhance narrative quality, consistency, and defensibility while saving up to $300,000 operationally. With the review time being cut by roughly 80%, AHT time has also been reduced, lightening analysts' burden and allowing them to spend more time on the alerts that require investigation.

Amazon Bedrock is facilitating secure data handling, encrypting all data, and allowing users to customize models privately. Integrated with riskCanvas, this powerful combination is enabling customers to enhance productivity in investigating, detecting, and preventing financial crime threats.

Plans are underway to include additional testing and potential implementation of STP for lower-risk screening events, which could lead to even greater savings of time and money. The success of this particular gen AI use case is spurring its use in others, such as alert, case, and suspicious activity report (SAR) narratives.

Insights section