Credit Risk
1. Probability of Default (PD Model)
What
- 预测借款人违约概率(Logistic / XGBoost / NN)
Where(业务)
- Loan approval(批贷)
- Credit limit setting(额度)
Risk(必须会说)
- Underestimation → 放出坏客户 → 信用损失
- Overestimation → 拒绝好客户 → 收入损失
- Bias → 公平性问题(监管重点)
2. Loss Given Default (LGD) / EAD
What
- 违约损失率 / 风险暴露
Where
- Capital calculation(资本计提)
- IFRS9 / CECL
Risk
- 资本低估 → 监管违规
- 资本高估 → ROE下降
3. Credit Scoring(零售信用)
What
- 客户评分(FICO-like)
Where
- 信用卡审批
- 消费贷
Risk
- Discrimination risk(种族/性别)
- Explainability不足(监管 challenge)
Market Risk / Trading
4. Pricing Models
What
- 衍生品定价(Black-Scholes / ML)
Where
- Trading desk
Risk
- Mispricing → 直接 PnL损失
- Model instability → 市场波动放大
5. VaR / Risk Models
What
- Value at Risk
Where
- Risk limit setting
Risk
- 低估风险 → 爆仓风险
- 模型假设失效(极端市场)
6. Algorithmic Trading / Signal Models
What
- ML预测价格走势
Where
- 自动交易系统
Risk
- Overfitting → 实盘崩溃
- Feedback loop → 市场冲击
三、Fraud / AML(非常常见 AI 场景)
7. Fraud Detection(反欺诈)
What
- 检测异常交易(ML / anomaly detection)
Where
- 实时交易监控
Risk
- False negative → 欺诈损失
- False positive → 客户体验差
8. AML Transaction Monitoring
What
- 检测洗钱行为
Where
- 合规系统
Risk
- 未识别洗钱 → 监管罚款
- 误报过多 → 运营成本高
四、Customer / Marketing(经常被忽略)
9. Customer Segmentation / Recommendation
What
- 客户分群 / 推荐产品
Where
- Marketing / cross-sell
Risk
- 错误推荐 → 收益下降
- 数据隐私风险
10. Churn Prediction(客户流失)
What
- 预测客户是否流失
Where
- retention strategy
Risk
- 错误预测 → 错失客户
五、Operations / Process Automation
11. Document Processing(OCR + NLP)
What
- 自动读取贷款文件
Where
- Loan processing
Risk
- 信息提取错误 → 决策错误
12. Credit Underwriting Automation(AI审批)
What
- 自动审批系统
Where
- End-to-end lending
Risk
- 黑盒决策 → 监管风险
- 无法解释拒贷原因
六、GenAI(现在最关键的未来方向)
13. Customer Service Chatbot(LLM)
What
- AI客服
Where
- 客户交互
Risk
- Hallucination → 错误信息
- 不一致回答
14. Internal Copilot(员工辅助)
What
- 帮分析报告 / 写代码
Where
- 内部效率工具
Risk
- 错误建议 → 决策风险
- 数据泄露
15. AI for Credit Decision Support(重点)
What
- LLM辅助信用分析
Where
- Analyst decision support
Risk
- 不可解释
- 决策责任不清
七、你面试时的“黄金用法”(重点)
你不需要记全部,但要做到:
能随时拿2–3个场景展开
推荐你优先准备:
1. PD Model(你主战场)
2. Fraud Detection(AI典型)
3. GenAI(拉开差距)
八、给你一个“万能回答模板”
当面试官问:
“Give me an example of AI model usage in banking”
你可以这样答:
In banking, AI models are widely used across different functions.
For example, in credit risk, machine learning models are used to predict probability of default and support lending decisions.
In fraud detection, AI models monitor transactions in real time to identify suspicious activities.
More recently, generative AI is being used in customer service and internal decision support.
Each of these applications introduces specific risks, including financial loss, regulatory exposure, and reputational damage.
