Subscribe to Tendercast, Data Science Statistics, Data Preparation, Machine Learning with Python courses to receive complimentary 4-hour 1:1 use case support

Client Risk Rating (CRR), Customer Due Diligence (CDD), Secrecy Act, and Screening with Machine Learning

Client Risk Rating (CRR) and Screening with Machine Learning is a perfect approach to obtain a preliminary risk score. Employing machine learning can support financial institutions to meet the Customer Due Diligence (CDD) rule and assess the risk of customers.  

CDD's four core requirements are the foundational tenets, however with the approach by which and augmentation of these tenets can be enhanced using machine learning.  

To review CDD's four requirements:  

  1. identify and verify the identity of customers
  2. identify and verify the identity of the beneficial owners of companies opening accounts
  3. understand the nature and purpose of customer relationships to develop customer risk profiles
  4. conduct ongoing monitoring to identify and report suspicious transactions and, on a risk basis, to maintain and update customer information

Employing a machine learning algorithm for new, and of course existing, clients can create a risk score and ongoing screening.  Machine learning network analytics using feature importance, network analytics and predictive risk scores based on logistic regression and random forest are great ongoing tools to produce a risk score.  Data, internal and external, is of course paramount which can be augmented by using machine learning's natural language processing (NLP) of both sanctions, subpoenas and news releases.  Maintaining and building a dictionary for NLP will only enable the cross utilization of analysis to extend across the bank, data sets and be employed to client data.  Combining the internal and external analysis and association of risk scores using machine learning provides a risk score for new clients as well as monitoring of existing.  

Please see the other article on integration and interpretation of risk scores across the algorithms.