Applying Machine Learning for Effective Customer Risk Assessment

This chapter focuses on three specific topics from a customer risk perspective—KYC process, watchlist screening, and customer risk assessment. While a lot of financial institutions are already working on automating the KYC process, authors have tried to provide an end-to-end process of digitization using application of OCR, computer vision, and basic machine learning-driven algorithms. Watchlist screening is a slightly more complicated topic requiring a robust machine learning-driven name-matching algorithm, where additional information can further reduce false positives. A detailed process of name matching is discussed that starts from preprocessing of names, transliteration and standardization of names and eventually name-matching algorithms. Authors have also provided an overview of a few prominent name-matching algorithms that focus on phonetics and edit distance. Eventually, the process of ensembling and machine learning was discussed that provide the best possible results. The last section in the chapter discusses customer due diligence process. It explains risk assessment scorecards. Authors have provided overview of expert-based and machine learning-driven scorecard. The chapter provides detailed explanation of machine learning-driven modeling process going upto discussion on the factors that are typically used in the development of such a scorecard.