The client needed to complete a time-sensitive 18-month lookback that was expected to take months of full-time staff at a substantial cost, and varying quality using traditional means.
Data Cleansing: The original data provided by the bank was incomplete and noisy. To separate noise and useful information, we undertook a cleansing exercise prior to starting the analysis. Using SQL and Python, we applied advanced analytics, pattern analysis, and indicator analysis to organize, deduplicate and consolidate the data, which included customer, payments, and transaction monitoring data.
Transaction Analysis Using Machine Learning Models: After developing product-specific scenarios using our industry expertise, we leveraged machine learning to rank data in certain categories and applied a random forest model to conduct preliminary reviews. We incorporated entity-level diligence information using DDIQ as a model input to enhance associated party risk considerations.
Analytics-Driven Reviews of Unstructured Trade Data: To review more than 3,000 trade finance transactions for red flags as part of the lookback, we leveraged OCR and pattern recognition technology to help screen information within nearly 100,000 associated documents for sanctions hits and trade finance red flags.
Manual Adjudication Supplement: Finally, we used our team to perform a manual review on the escalated transactions and documents containing red flags.
The data cleansing effort consolidated data from disparate systems to compile more than 70,000 in-scope transactional histories and more than 30,000 unique entities. DDIQ enhanced entity-level diligence information, enriched transactional data and conducted a preliminary review to significantly reduce the case population for manual review. With a well-documented methodology for automated review, the manual review consisted only of a limited population of alerts consisting of 10,000 transactions and 800 entities, accelerating completion while maintaining high quality levels.