Building Comprehensive Risk Feature Portraits for All Loan Stages

Risk control feature portraits describe customer risk from multiple dimensions. By subdividing dimensions, we can accurately characterize each specific aspect for precise risk identification.

The portrait system can be applied to four scenarios—marketing, pre‑loan, in‑loan, and post‑loan—by selecting suitable data dimensions for each.

Marketing Feature Portrait

For historical customers, data includes basic information, historical application records, and multi‑loan information. For new customers, data may consist of browsing behavior, partial basic info, and third‑party data. The main data dimensions and derived feature portraits are illustrated in the accompanying diagrams.

Pre‑Loan Feature Portrait

Used for anti‑fraud, credit risk assessment, and pricing. Key considerations are repayment ability and compliance probability, derived from basic income, assets, historical loans, social relations, and behavior habits. Data sources include self‑filled information, authorized data, and external integrations.

Key data categories:

Customer Basic Information : self‑filled registration data and credit report basics, covering personal, employment, income, family, loan, public service, and address details.

Customer Authorized Data : data obtained with consent, such as device fingerprints, contacts, app lists, and GPS, reflecting behavior and social ties.

Customer Behavior Data : app and server event logs, consumption, and web browsing, describing usage preferences.

ID‑Linked Data : multiple identifiers (ID card, phone, device) revealing hidden relationships and enabling cross‑reference of orders.

Historical Order Data : past order compliance or default records, analyzed by identity, order type, contacts, GPS, and amount metrics.

In‑Loan Feature Portrait

Uses all pre‑loan data (with possible value changes) plus current unfinished orders, in‑loan behavior events, approval results, and repayment reminders. These dimensions capture how customer needs evolve after the first disbursement.

In‑Loan behavior and order data: illustrated with visual examples.

Customer authorized data and its changes: variations may signal shifting default risk.

Post‑Loan Feature Portrait

Reflects post‑loan default risk, focusing on repayment willingness and ability. Utilizes historical order outcomes, post‑loan follow‑up records, post‑loan behavior data, and changes across all previous stages.

Data dimensions are split into two categories: (1) post‑loan behavior, approval results, and follow‑up records; (2) information changes compared to pre‑loan and in‑loan stages. The generation method mirrors that of pre‑loan and in‑loan portraits.