1. Architecture of Non-Performing Loan Management System
The architecture of the non-performing loan management system consists of four layers, from bottom to top (see Figure 1).
First Layer: Data Source Layer. Responsible for data collection. Firstly, it connects to the loan transferor's bank credit system to obtain basic information about the borrower. Then, it integrates third-party data sources to access the borrower's authentic data. Lastly, it connects to judicial litigation data to ascertain the borrower's litigation status.
Second Layer: Data Platform Layer. This layer establishes a data marketplace based on collected data, forming a data lake and a data warehouse, which includes sections like Customer Marketplace, Asset Marketplace, Financial Marketplace, and Public Marketplace.
Third Layer: Data Application Layer. This layer includes essential functional modules like Due Diligence, Panoramic Profiling, Intelligent Assessment, Intelligent Collection, External Collection, Smart Court, Judicial Auction, Smart Operations, etc., enabling automatic business processes using AI and machine learning.
Fourth Layer: Business Layer. The main business modules of the asset management company, focusing on corporate non-performing loans and personal non-performing loans. The evaluation of corporate non-performing loans mainly emphasizes collateral assessment, differing from personal non-performing loans.
2. Non-Performing Loan Disposal Process
For the disposal of personal non-performing loans, the whole process management ranges from data collection to repayment, mainly in four stages (see Figure 2).
First Stage: Data Collection. Interface with the transferor's credit system to obtain the basic loan data, integrate identity data via third-party platforms, and gather the borrower's litigation data via judicial interfaces.
Second Stage: Due Diligence. Establish a borrower's customer profile and conduct automated due diligence. With a comprehensive understanding of the borrower's credit status, employ AI for intelligent assessment to provide value appraisals and opinions.
Third Stage: Collection. First, use email, voice, and other means to collect loans. For most personal non-performing loans, use external collection, litigation collection, etc., to achieve clearance.
Fourth Stage: Repayment. Achieve repayment accounting and finalization.
3. Explanation of Some Core Links
(1) Intelligent Assessment
Traditionally, the assessment of corporate non-performing loans places a strong emphasis on collateral valuation or the evaluation of a company's debt repayment capability. This is an area where most asset management companies excel. However, such evaluations for individual non-performing loans aren't suitable for qualitative expert assessments. Instead, technologies such as big data, artificial intelligence, and machine learning can be used for intelligent evaluations. This reduces manual costs, enhances the accuracy of individual non-performing loan evaluations, lowers the evaluation difficulty, and assists in improving the recovery ability, thus paving the way for new arenas in the bulk transfer of individual non-performing loans.
Intelligent assessment mainly analyzes foundational loan data provided by the assignor, user behavioral data, and third-party authorized data. Basic data includes not only public information like age and account duration from the Banking Industry Credit Asset Registration and Circulation Center but also non-public details such as the borrower's occupation, job duration, annual income, housing situation, etc. User behavior data encompasses historical loan amounts and overdue repayment amounts. To determine the precise investment value of asset valuation, modeling of external data is vital, requiring third-party credit data to complement user credit information. The authenticity of this data must be verified from third parties. For example, information such as bank card verification, ID card verification, phone number verification, blacklists of defaulters, and comprehensive court litigation data can be acquired from the JD Data website. Personal online loan data can be fetched from websites like Sesame Credit, Tencent Credit, and Qianhai Credit. Websites like Yunfang Data provide valuation data for collaterals like real estate. Integrating internal and external data can enhance evaluation accuracy.
Based on this, data is settled to form data assets. Utilizing the recovery prediction model, AI techniques like logistic regression, decision tree regression, cluster analysis, and relationship graphs can be employed to analyze and evaluate asset recovery rates, intelligently matching solutions to enhance the efficiency of non-performing loan management.
Considering issues related to user privacy, it is essential to anonymize user data or adopt privacy computing methods to ensure the safety of user data.
(2) Intelligent Collection
Traditional collection tasks are inefficient, with unregulated business management and an unmonitored collection process. Intelligent collection systems often include collection work systems, voice call systems, integrated business management systems, and intelligent service systems. However, while intelligent collection is effective for short-term loans, especially those less than 2 months old, it's not advantageous compared to manual collection for longer-duration loans.
