Risk Control and Operations for Existing Credit Customers: Models, Strategies, and Practices

The presentation focuses on risk control and operational strategies for existing (stock) loan customers, highlighting the impact of COVID‑19‑driven client migration, evolving regulations, and faster deterioration cycles.

Key points include: (1) accelerated client flow among P2P, small‑loan, consumer finance, and banking institutions; (2) regulatory changes causing short‑term fraud risk spikes; (3) reduced warning‑to‑deterioration time from 1‑2 months to 1‑2 weeks; (4) increased segmentation and operational pressure due to pricing standards.

1. Existing Customer Risk Warning Process – Combines rule‑based and model‑based alerts with three strategy layers: strong rules (e.g., police, court data), weak rules (monitoring signals), and dynamic rule switching to balance risk detection and customer experience.

2. Full‑Coverage B‑Card Strategy – Uses rule‑based and model‑based segmentation (low, medium, high risk) and applies targeted actions such as account freezing, limit reduction, interest adjustment, or cross‑selling based on usage patterns and debt levels.

3. Behavior Models

Two B‑Card variants are described: classic B‑Card (monthly batch, based on historical repayment data) and high‑frequency B‑Card (daily/weekly batch using app event data). The high‑frequency model improves detection of rapidly deteriorating customers.

3.1 LSTM – Long Short‑Term Memory networks capture time‑series features to enhance predictive power over traditional scorecards, typically improving KS by 3‑5 points.

3.2 Survival Model – Adapts medical survival analysis to credit risk, estimating the probability of default over specific future horizons, enabling time‑based risk forecasting.

4. Existing Customer Operations – Describes a three‑layer framework (backend data hub, middle‑office analytics, frontend engagement) to build a value‑enhancement system, conduct deep customer analysis, and execute cross‑selling via H5, calls, app recommendations, and offline events.

5. Capability Map – Highlights five core abilities: data collection, feature engineering, data mining (marketing/decision models), decision tools (graph models, BI platforms), and data application (lifecycle management, intelligent campaigns).

6. Customer Segmentation and Lifecycle – Defines operable categories (potential, new, stable, dormant, lost) and outlines strategies to convert prospects, retain stable users, reactivate dormant accounts, and prevent churn.

7. Marketing Model Process – Differentiates between old and new customers, builds response models, and balances marketing conversion with credit risk, emphasizing precise targeting to avoid inverse selection.

8. Model Evaluation – Advocates for a “deal‑closing” model that jointly optimizes response rate and credit risk, and suggests combining B‑Card screening with marketing models to create a whitelist for higher‑quality outreach.

In summary, the article provides a comprehensive framework for risk monitoring, advanced modeling (LSTM, survival analysis), and operational tactics to enhance the value and retention of existing loan customers while managing credit risk.