Customer credit risk assessment is crucial in retail banking business. Hitherto, customer credit risk is narrowed to the direct loss resulting from the borrower’s failure to contractual obligation of loan products. The indirect loss, such as legal and compliance liability, reputation damage, is usually neglected. The indirect loss is generally caused by providing services to customers who launch illegal transactions, such as money laundering and terrorist financing activities. As required by Basel Committee on Banking Supervision, banks should strengthen the customer risk supervision and avoid providing services to abnormal customers. Usually, a customer needs to apply for an account before launching financial activities. Thus, a risk assessment at the account opening services can not only help the bank filter some fraudulent and criminal customers but also provide guidance on allocating the limited supervision resource. Conventionally, such a risk assessment procedure is labor-intensive. Given the massive business volume caused by digitalization, the problems of human resource shortage and high human errors caused by overburdened workload are even more serious. Yet, previous works only explored abnormal account activities supervision after account generation, whereas the automatic risk assessment method on account opening services is rarely covered. To address these problems, in this study, a novel preliminary risk level is firstly developed to indicate the different likelihood of risk levels in account opening service. In addition, an automatic risk level classifier is developed with a designed two-staged neural network utilizing geographic, demographic, and behavioral features. The experiment results verify the high classification accuracy at 96.7%. Moreover, given the similar compositions of international customers between Korea and Hong Kong, this work could also provide guidance to Korean banks on customer preliminary risk assessment.
