DRIVING EFFICIENCY IN SANCTIONS SCREENING
Sanction screening weakness can pose a significant problem for financial institutions and their customers. It’s a critical compliance process that helps identify and prevent financial transactions involving individuals, entities, or countries subject to economic sanctions.
When an organisation has a sanction screening weakness, it means that its processes for identifying sanctioned entities or individuals may not be effective, leaving the institution vulnerable to financial crime, including money laundering and terrorist financing. This can lead to heavy fines, loss of business and reputational damage. Moreover, customers may also be affected if they inadvertently engage in transactions with sanctioned individuals or entities, potentially facing legal and reputational consequences themselves.
The consequences of poor sanction screening
These were exactly the issues faced by a leading European Private Bank. The weaknesses related to the architectural set up of the vendor system and a partial outsourcing of scoring to a third party. They required a new solution and tuning support to optimise and benchmark the system.
A key objective was to lower the bank’s false positive rate. The false positive rate refers to the percentage of flagged transactions that are determined to be false positives, or transactions that are wrongly flagged as potential sanctions violations. In other words, it’s the ratio of the number of transactions that are flagged for further review to the number of transactions that are sanctioned. A high false positive rate can be a significant problem as it can lead to unnecessary delays in processing legitimate transactions, increased costs and decreased efficiency. Moreover, if a bank’s false positive rate is too high, it can lead to compliance officers overlooking true violations, which can result in regulatory fines, reputational damage and legal consequences. Therefore, it’s essential for banks to maintain a low false positive rate to ensure that they can accurately identify and prevent financial crimes while minimising disruptions to legitimate transactions.
An initial capability assessment was used to identify the weaknesses and provide recommendations to improve the system’s effectiveness. This resulted in a re-assessment of the institution’s risk profile and enabled the bank to reduce its false positive rate by 6.4%. In addition, the recommendations also enabled the bank to decrease the number of hits the system returned by 50% without impacting its ability to identify true matches, resulting in greater system effectiveness and increased performance which ultimately reduced overall screening costs. To further reduce the false positive rate, machine learning algorithms were applied to the post-processor. These were trained on large datasets of known sanctioned and non-sanctioned individuals and entities to learn patterns and identify features that are indicative of sanctions. The post-processor refined the results of the initial screening process to improve its accuracy, resulting in the number of false positives falling by a further 11%.
One of the recommendations also highlighted in the initial capability assessment was the tuning of date of birth parameters. Our experience revealed that often this parameter was set to the default setting. Sanction screening systems typically use date of birth as one of the key criteria for identifying and flagging sanctioned individuals or entities. However, the accuracy of this criterion can be significantly impacted by various factors, such as variations in date format and data quality issues.
Date of birth tuning involves adjusting the format of the date of birth, such as changing the order of the day, month and year, or specifying a range of acceptable values for each component. This ensures that the system can correctly interpret and match the date of birth information provided in the screening data. Another approach is to apply fuzzy matching techniques that account for variations in data quality and format by using algorithms to identify and compare partial matches and variations in the date of birth information, such as missing or incorrect digits. Our approach altered the variable by calculating the difference in date of birth and the assigned weight in the aggregation of the final match score to improve the system’s performance. The optimisation regime used the output dataset and considered over 4 million total score options for each of the returns. Based on the system’s existing threshold, a combination of parameters were investigated to introduce a scoring function that had the highest impact on suppressing false positives, but with minimal impact on reducing the system’s ability to capture sanctioned entities. Once an appropriate configuration was deduced, comparisons of the performance against other solutions were made to confirm the stability of the recommendation. The combination yielded a significant and stable reduction of 10%, with no impact to control and manipulated records.
Due to the success of the above projects, we also worked collaboratively with the bank to provide assurance of their adverse media system. An adverse media system is a tool used in sanctions screening to identify any negative news or information about an individual or entity that may be indicative of financial crime or other illegal activities. This system uses advanced algorithms and natural language processing techniques to scan large volumes of news articles, social media posts, and other public records to identify any information that may be relevant to a sanctions screening process.
The adverse media system is an essential component of sanctions compliance programs, helping to identify potential risks and ensure that they are not inadvertently doing business with individuals or entities that are subject to sanctions. The system will flag potential matches for further investigation by the compliance team, who then determine whether a transaction or relationship should be allowed to proceed. It is therefore very important to get the balance right as too high and more flags will be raised meaning more time will be spent having to manually investigate risks, whilst too low and the bank runs the risk of falling foul of AML regulations.
A statistical analysis of the cap the bank used to determine the volume of adverse media documents was run to determine if the cap was fit for purpose. It also explored other potential cap settings to ensure compliance to regulations and to optimise the number of documents being generated in relation to the risk appetite of the bank.
The goal of the analysis was to estimate the probability that for a specified number of documents at least one document is relevant. This was achieved by implementing a customised analytical framework for estimating the marginal probability of documents in a row to understand their average relevance. To minimise bias and ensure the accuracy of the results the relevance was determined based on the frequency of languages per data vendor for each territory and thus the findings deduced illustrated the confidence level and hence the stability of the results. The analysis proved that for all data vendors the existing cap was appropriate to reduce relevant results, while minimising false positive without missing identified true alerts. It was also found in some cases that it would be appropriate to implement a marginal reduction in the cap for each data provider with minimal impact (less that 1 per cent) to the existing system performance, further demonstrating the reliability of the existing cap.
Clearly, the application of data science and machine learning techniques are important in sanctions screening and the maintenance of AML compliance. They enable organisations to analyse large volumes of data more efficiently and accurately than would be possible using traditional methods. As these activities demonstrate optimising the process of data analysis and more accurately flagging potential risks, means suspicious activity can be identified more quickly, more accurately, more effectively and more efficiently.
Discover how Beyond can empower your business with unparalleled insights through our comprehensive data analytics solutions. Contact us today to begin your journey towards enhanced business insight and informed decision-making.
Enjoyed this article? Try The Advantages of Credit Risk Models.