Case Investigation Pattern(案件调查模式)
知识库位置:
FDE Knowledge Base/12 Patterns/05-Case-Investigation-Pattern.md
Problem Signature(问题签名)
Section titled “Problem Signature(问题签名)”专业人员需要调查一个复杂 Case(案件),信息散布在多个系统。
常见于:
- Fraud(欺诈)
- AML(Anti-Money Laundering,反洗钱)
- Security(安全)
- Claims(理赔)
- Compliance(合规)
- Customer Escalation(客户升级投诉)
- Incident Investigation(事件调查)
Current Workflow(现状工作流)
Section titled “Current Workflow(现状工作流)”Case Created(立案)↓Read Case(读案件)↓Search System A(搜系统 A)↓Search System B(搜系统 B)↓Search Documents(搜文档)↓Build Timeline(构建时间线)↓Compare Policy(对照政策)↓Reach Conclusion(形成结论)↓Document Findings(记录调查结果)Pain(痛点)
Section titled “Pain(痛点)”- Search time(搜索时间)高
- Evidence(证据)分散
- Investigation quality varies(调查质量不稳定)
- Documentation slow(记录缓慢)
- Expert dependency(依赖专家)
Target Workflow(目标工作流)
Section titled “Target Workflow(目标工作流)”Case(案件)↓Investigation Workspace(调查工作台)├─ Structured Data(结构化数据)├─ Documents(文档)├─ Event Timeline(事件时间线)└─ Policy Knowledge(政策知识)↓AI Copilot(AI 副手)↓Evidence Summary(证据摘要)↓Suggested Investigation Steps(建议调查步骤)↓Human Investigator(人工调查员)↓Decision(决策)↓Audit(审计)Architecture(架构)
Section titled “Architecture(架构)”组合:
AI Copilot(AI 副手)+Enterprise RAG(企业级检索增强生成)+API Integration(接口集成)+Human-in-the-Loop(人在环)必要时增加 Agent(智能体)做受控信息采集。
AI Role(AI 角色)
Section titled “AI Role(AI 角色)”- Summarization(摘要)
- Evidence synthesis(证据综合)
- Timeline generation(时间线生成)
- Knowledge retrieval(知识检索)
- Suggested next investigation step(建议下一步调查动作)
What AI Should NOT Own(AI 不应负责的部分)
Section titled “What AI Should NOT Own(AI 不应负责的部分)”通常不建议模型独立完成:
- Fraud disposition(欺诈定性)
- Regulatory decision(监管决策)
- High-impact account closure(高影响账户关停)
- Final legal conclusion(最终法律结论)
Key Design Principle(关键设计原则)
Section titled “Key Design Principle(关键设计原则)”系统必须区分:
Fact(事实)Evidence(证据)Inference(推论)Recommendation(建议)Decision(决策)不要让 AI 生成内容混在一起。
Required Data(所需数据)
Section titled “Required Data(所需数据)”- Case record(案件记录)
- Customer / entity data(客户/实体数据)
- Transactions / events(交易/事件)
- Documents(文档)
- Policy(政策)
- Historical cases(历史案件)
Security(安全)
Section titled “Security(安全)”通常非常高。
需要:
- Strict authorization(严格授权)
- Evidence lineage(证据溯源链)
- Immutable audit(不可篡改审计)
- Sensitive data handling(敏感数据处理)
Pilot(试点设计)
Section titled “Pilot(试点设计)”建议:
Historical closed cases(历史已结案件)↓Replay(回放)↓Compare AI-assisted workflow(对比 AI 辅助流程)vsoriginal investigator workflow(原调查员流程)再进入 Live Pilot(真实试点)。
Metrics(指标)
Section titled “Metrics(指标)”- Investigation time(调查时长)
- Evidence coverage(证据覆盖率)
- Analyst acceptance(分析员接受度)
- Escalation(升级)
- False positive / negative(误报/漏报,如适用)
- Documentation time(记录时间)
Failure Modes(失败模式)
Section titled “Failure Modes(失败模式)”- AI invents evidence(AI 编造证据)
- No source lineage(没有来源溯源)
- Investigation bias(调查偏倚)
- Wrong entity matching(实体匹配错误)
- Excessive automation(过度自动化)
Industry Mapping(行业映射)
Section titled “Industry Mapping(行业映射)”尤其适合:
- Financial Services(金融服务)
- Insurance(保险)
- Security Operations(安全运营)
- Compliance(合规)