pre-visit-credit-analysis

Pre-visit Penetrating Analysis

Before visiting a customer, conduct a systematic background investigation and risk screening of the customer to ensure that visit objectives are clear and risks are known.

Target Role

Role: Corporate account manager, risk manager, credit approval officer
Use Cases: Background investigation before the first visit to a new customer, risk assessment before renewal for existing customers, visit preparation for non-performing early-warning customers, pre-visit reconnaissance for large new credit grants
Output Purpose: Generate a structured "pre-visit one-pager" for formulating visit plans, preparing a list of risk questions, and determining key visit directions
Decision Level: Triggering a one-vote veto (V1-V6) must be reported to the risk management department; red-flag signals must be verified with priority during the visit; yellow-flag signals are included in routine attention
Execution Frequency: Mandatory before every on-site visit; mandatory for a new customer's first credit grant/renewal for existing customers/triggering of risk early warning

Data Sources

Required Data

Data Item	Source	Acquisition Method	Sensitivity Level
Corporate business registration data	Qichacha/Tianyancha/National Enterprise Credit Information Publicity System	API interface or MCP tool	Public
Judicial risk data	Ping An Credit/Qichacha judicial module/court announcement website	API interface or MCP tool	Public
Public opinion monitoring data	Bailian Web Search/news public opinion system	Web search or MCP tool	Public
In-bank deposit and loan data	CRM system/core banking system	Internal API or MCP tool	Internal
External credit reporting data	People's Bank of China credit reporting system	Internal API (requires written customer authorization)	Sensitive

Data Desensitization Rules

Customer names and unified social credit codes use desensitized placeholders in the output (e.g., "XX Manufacturing Co., Ltd.")
Personal ID numbers and bank account numbers in credit reports must be replaced with[证件号码]、[银行账号]
Specific amounts in in-bank deposit and loan data may use interval descriptions (e.g., "10-20 million")

Degradation Strategy

Business registration data interface unavailable: Use web queries on the National Enterprise Credit Information Publicity System as a fallback, and mark "Data source: web query, timeliness may be delayed"
In-bank CRM system unavailable: Skip in-bank data analysis, clearly mark in the report "In-bank data unavailable, recommend requesting it from the customer during the visit", and continue executing other steps
Credit reporting system unavailable or authorization not obtained: Record "Credit authorization not obtained, skipping credit inquiry", rely only on public data (business registration/judicial/public opinion), and note the limitations in the report
Public opinion search tool unavailable: Use a general search engine (Web Search) as a substitute, expand the keyword scope, and note "Public opinion data source: general search, coverage may be limited"

Constraints

Regulatory basis: "Administrative Measures for Commercial Bank Loan Business" (CBRC Order No. 2 of 2010) — pre-loan investigation requirements "Notice on Strengthening Loan Risk Management" (Yin Jian Fa [2020] No. 22) — risk identification and early warning "Enterprise Credit Report" inquiry compliance requirements — credit inquiries must obtain written authorization Bank internal pre-loan investigation operating procedures — specific pre-visit analysis standards

Data source transparency: All data points must indicate the source tool/channel; data must not be fabricated or inferred
Authorization compliance: Inquiring credit reports requires confirming that the customer's written authorization has been obtained; otherwise, only public information may be checked
Red flags first: If a veto signal (V1-V6) is found, it must be prominently marked immediately and must not be downplayed in the summary
Tool failure does not interrupt: When a data source call fails, record the reason and continue with other steps; analysis must not be stopped because of a single failure
Objective and neutral: Present data truthfully; conclusions must be supported by evidence and must not involve excessive subjective inference
Skipping steps prohibited: Step 0 (data confirmation) and Step 6 (veto self-check) must not be skipped; the complete process must be executed
Unauthorized recommendations prohibited: Output only pre-visit risk analysis; do not provide credit approval decisions, loan pricing recommendations, or risk classification adjustment recommendations

Workflow

📋 Interaction mode: Mode A - Report Generation Regulatory basis: "Administrative Measures for Commercial Bank Loan Business" (CBRC Order No. 2 of 2010) — pre-loan investigation requirements

Step 0: Data Confirmation and Verification (read before write)

Execution instructions: Before starting the pre-visit analysis, the completeness and validity of the input data must be confirmed.

Execution steps:

Confirm that the customer name or customer ID is non-empty and that the Unified Social Credit Code (if any) is in the correct format
Confirm the purpose of the visit (first loan for a new customer/renewal of an existing loan/non-performing early warning/large new addition), which affects the focus of the analysis
Confirm the credit authorization status (obtained/not obtained) to determine whether to execute Step 5
Run scripts/validate_pre_visit.py verify the input data format and required fields
Only after verification passes may Step 1 be entered; if verification fails, return the list of missing data

⚠️ Required field checks must not be skipped; the customer name/ID and visit purpose must be confirmed as complete

Step 1: Collection of Basic Enterprise Business Registration Information

Execution instructions: Obtain the enterprise's business registration information and penetrate the equity structure down to the ultimate natural person.

Execution steps:

Query the enterprise's business registration information: full name, Unified Social Credit Code, legal representative, registered capital, date of establishment, business scope, shareholder structure
Verify the paid-in ratio of registered capital: paid-in ratio < 30% → mark ⚠️ yellow flag
Verify whether the business scope covers the industry corresponding to the applied business: mismatch → mark ⚠️ yellow flag
Perform equity penetration (at least 3 layers) to identify the actual controller: penetration beyond 3 layers → mark ⚠️ yellow flag and pay attention to nominee holding risk
Check whether the legal representative/actual controller overlaps with known risk lists: overlap → mark 🚩 red flag

⚠️ The equity penetration step must not be skipped; penetration must go through to the ultimate natural person, and must not stop at intermediate legal entities

Step 2: Judicial risk screening

Execution instructions: Obtain the enterprise's judicial risk records and identify major litigation and administrative penalties.

Execution steps:

Check records of dishonest judgment debtors: if present → trigger V1 veto
Check judgment documents: amount involved > 30% of registered capital → mark 🚩 red flag
Check administrative penalty records in the past 2 years: ≥ 3 times → mark 🚩 red flag
Check abnormal business status: if the status has not been lifted → mark ⚠️ yellow flag, require on-site verification
Check personal litigation of the legal representative/actual controller: amount > 500,000 → mark ⚠️ yellow flag

⚠️ The dishonest judgment debtor check must not be skipped; the V1 trigger conditions must be verified one by one

Step 3: Business status and public opinion scanning

Execution instructions: Obtain the enterprise's business dynamics and public opinion information, and identify negative signals.

Execution steps:

Execute public opinion search, keywords:企业名称 + 违规/欺诈/跑路/资金链/债务危机
Focus on public opinion in the past 6 months: if negative → mark 🚩 red flag; historical negative information older than 1 year → mark ⚠️ for reference
Check outbound invested enterprises: if there are dishonest/cancellation records → mark ⚠️ yellow flag
Check honors and qualifications (high-tech enterprise, specialized and sophisticated enterprise, etc.): if within the validity period → record as ✅ positive signal

⚠️ Must not search only a single keyword; must cover at least 5 risk keyword combinations

Step 4: In-bank deposit and loan data collection (if available)

Execution instructions: Obtain the customer's in-bank deposit and loan records and credit occupation status.

Execution steps:

Check the customer panorama view: deposit balance, credit balance, loan details
Calculate credit occupation rate: occupation rate > 90% → mark ⚠️ yellow flag, pay attention to whether there is room for new demand
Check maturity reminders: credit/loans maturing within 90 days → mark ⚠️ yellow flag, prepare loan renewal materials in advance
Check historical overdue records: overdue in the past 2 years → mark ⚠️ yellow flag, understand the reasons
Calculate deposit coverage multiple: deposit balance < 2 times the average monthly repayment amount → mark ⚠️ yellow flag, pay attention to liquidity

⚠️ If the in-bank system is unavailable, record "in-bank data unavailable", continue executing other steps, and do not interrupt

Step 5: External credit reporting collection (customer authorization required)

Execution instructions: Obtain the People's Bank of China credit report and analyze credit history.

Execution steps:

Confirm credit authorization status: if authorization has not been obtained → record "authorization not obtained, skip credit inquiry", enter Step 6
Check credit report: credit inquiry count in the last 6 months > 6 times → mark ⚠️ yellow flag (signal of frequent borrowing)
Check external guarantee situation: total external guarantees > 50% of net assets → mark ⚠️ yellow flag
Check outstanding overdue loans: existence of M3 or higher overdue (within 6 months) → trigger V2 veto

⚠️ If credit authorization cannot be obtained, it must be explicitly recorded and continue; the analysis must not be suspended due to lack of credit data

Step 6: Veto condition self-check and report generation

Execution instructions: Compare V1-V6 veto conditions and generate a structured "pre-visit one-pager".

Execution steps:

Compare the V1-V6 veto conditions one by one:
- V1: The enterprise or the actual controller is included in the list of dishonest judgment debtors and has not been removed
- V2: There are unsettled overdue loans in the credit reporting system (M3 and above, within 6 months)
- V3: The enterprise has been included in the list of business abnormalities for more than 1 year, and there is no record of removal
- V4: Within the past 2 years, there are records of administrative penalties or criminal case filing related to suspected financial fraud or money laundering
- V5: The industry to which the enterprise belongs is subject to explicit regulatory restriction/elimination (the elimination category in the Catalogue for Guiding Industry Restructuring)
- V6: The actual controller has recently fled or become unreachable, or there is related negative public opinion that has been cross-verified as true through multiple sources
If any veto condition is triggered, mark it in red at the top of the report: [Veto-VX: Trigger Reason]
Generate a structured "pre-visit one-pager", including enterprise overview, operating status, risk signal rating, key visit recommendations, and data sources
Call the note-saving capability to save the report to the customer notes area

⚠️ Do not omit any veto condition; V1-V6 must be compared one by one; if triggered, it must be marked in red at the top of the report and must not be downplayed

Output Format

Output a structured "pre-visit one-pager", including the following sections:

Section	Content	Data Type
Basic Report Information	Report number, generation time, customer name, visit purpose	Table
Enterprise Overview	Full name/unified social credit code/establishment date, legal representative/actual controller (penetrated to natural persons), registered capital (subscribed/paid-in), business scope, key points of equity structure	Table
Operating Status	Main business, recent operating developments (public opinion/announcements/investment trends), summary of in-bank deposit and loan data (deposit balance/credit balance/utilization rate/recent repayments), qualifications and honors (including validity period)	Table
Risk Signal Rating	Red flag signals (major risks), yellow flag signals (potential risks), positive signals (factors supporting credit granting), listed one by one with data sources noted	List
Veto Check	V1-V6 comparison results one by one; if triggered, it must be marked in red	Table
Key Visit Recommendations	Matters that must be verified (targeting red/yellow flags), questions recommended to be understood, materials recommended to be collected	List
Data Sources	Status of the five major data dimensions (business registration/judicial/public opinion/in-bank/credit reporting), source tools/channels, remarks	Table

Report Template Example:

## 访前一页纸 — [客户名称]
> 分析时间:YYYY-MM-DD HH:MM
> 分析类型:访前穿透分析
> ⚠️ 如触发一票否决信号,在此处红色标注:【一票否决-VX:触发原因】

---

### 企业概况
- 全称 / 统一社会信用代码 / 成立时间
- 法人代表 / 实际控制人(穿透至自然人)
- 注册资本(认缴/实缴) / 经营范围
- 股权结构要点(如有异常穿透关系注明)

### 经营状况
- 主营业务 / 主要产品或服务
- 近期经营动态(舆情、公告、投资动向)
- 行内存贷数据摘要(存款余额、授信余额/占用率、近期还款情况)
- 资质荣誉(高新、专精特新等,含有效期)

### 风险信号评级
#### 🚩 红旗信号(重大风险,须决策是否继续推进)
- [逐条列出,注明数据来源和发现时间]

#### ⚠️ 黄旗信号(潜在风险,走访时重点核实)
- [逐条列出,注明核实建议]

#### ✅ 正面信号(支持授信的积极因素)
- [逐条列出]

### 走访重点建议
1. **必须核实的事项**:[针对红旗/黄旗信号的对应核实清单]
2. **建议了解的问题**:[需向企业方获取的信息,如财务数据、合同等]
3. **建议收集的材料**:[走访时应请企业提供的文件]

### 数据来源
| 数据维度 | 状态 | 来源工具/渠道 | 备注 |
|---------|------|-------------|------|
| 工商信息 | ✅成功 | XXX | - |
| 司法风险 | ✅成功 | XXX | - |
| 舆情 | ✅成功 | XXX | - |
| 行内数据 | ❌失败 | - | 原因:XXX |
| 征信 | ⏭️跳过 | - | 原因:未取得授权 |

After generation is complete, callnote-saving capability, and save the above "pre-visit one-pager" to the customer notes area. This output can be parsed and used by the credit-due-diligence and visit-memo Skills Reference shared/disclaimer-template.md template, and add a disclaimer at the end of the report

Audit Trail

After execution is complete, generate an audit log recording the following information:

{
  "skill_name": "pre-visit-credit-analysis",
  "execution_time": "2026-05-05T14:00:00+08:00",
  "customer_name": "[客户名称]",
  "visit_purpose": "新客户首笔/存量续贷/不良预警/大额新增",
  "data_sources": {
    "business_registry": {"status": "success", "source": "企查查API"},
    "judicial_risk": {"status": "success", "source": "凭安征信"},
    "public_opinion": {"status": "success", "source": "百炼Web搜索"},
    "internal_data": {"status": "failed", "reason": "CRM系统不可用"},
    "credit_report": {"status": "skipped", "reason": "未取得征信授权"}
  },
  "veto_checks": {
    "V1": false,
    "V2": false,
    "V3": false,
    "V4": false,
    "V5": false,
    "V6": false
  },
  "red_flags_count": 2,
  "yellow_flags_count": 3,
  "report_saved": true
}

Audit log retention period: at least 3 years.

Gotchas

#1: Equity penetration stops at an intermediate-level legal person

Symptom:The report only shows direct shareholders and does not penetrate through to the ultimate natural person actual controller, missing the risk of nominee shareholding.
Reason:The business registration data API returns limited levels, or the AI skips the deep penetration step.
Solution:Step 1 must enforce at least 3 levels of penetration; if there are more than 3 levels, mark "penetration chain too long, pay attention to nominee shareholding risk"; do not stop at an intermediate-level legal person.

#2: Single public opinion search keyword leads to omissions

Symptom:Only searching "enterprise name + violation" misses key negative information such as "capital chain rupture" and "actual controller unreachable".
Reason:Insufficient search keyword coverage; multi-dimensional combined searches were not performed.
Solution:Step 3 must cover at least 5 risk keyword combinations (violation/fraud/absconding/capital chain/debt crisis); do not use only a single keyword.

#3: Credit authorization not obtained but not explicitly recorded

Symptom:The report does not mention the credit inquiry status, so the user does not know whether a credit inquiry has been conducted.
Reason:When authorization has not been obtained, the AI directly skips Step 5 and does not explicitly record it in the report.
Solution:Step 5 must explicitly record "credit authorization not obtained, skipping credit inquiry" and mark "⏭️Skipped" in the data source table.

#4: Visit recommendations are too general and lack specificity

Symptom:Visit recommendations contain vague descriptions such as "understand the financial situation" and "verify operating conditions", which cannot guide an actual visit.
Reason:The AI did not translate risk signals into specific verification questions.
Solution:Visit recommendations must target the discovered red flag/yellow flag signals, such as "verify whether revenue over the past 2 years matches bank statements" and "ask about the current progress of litigation cases"; do not speak in generalities.

Examples

Example 1: Analysis before the first visit to a new customer

User input:

做一下XX制造有限公司的访前分析,首次走访,目的是新客户首笔授信,征信已取得授权。

Skill execution process:

Step 0: Confirm customer name "XX Manufacturing Co., Ltd.", visit purpose "first credit extension for a new customer", credit authorization "obtained"
Step 1: Query business registration information; found registered capital of 50 million, paid-in only 10 million (20%) → mark ⚠️yellow flag
Step 2: Query judicial risks; no dishonest execution records, but there is 1 contract dispute involving 8 million (16% of registered capital) → no red flag
Step 3: Public opinion search; found a news report from 3 months ago that "the company obtained high-tech enterprise certification" → record ✅positive signal
Step 4: Internal bank data unavailable (no records for a new customer) → record "internal bank data unavailable"
Step 5: Credit inquiry; no overdue records, 3 inquiries in the past 6 months → no yellow flag
Step 6: Compare V1-V6; no veto trigger, generate the "pre-visit one-pager"

Output points:

Enterprise overview: registered capital 50 million (paid-in 20%⚠️), established 2 years ago, legal representative Zhang San
Risk signals: 1 yellow flag (low paid-in ratio), 1 positive (high-tech enterprise)
Visit recommendations: verify the plan for paid-in capital to be in place, ask about the progress of the contract dispute, obtain financial statements

Example 2: Analysis before a visit for renewal of an existing customer's loan

User input:

查一下XX贸易有限公司的访前情况,存量客户续贷,走访时间下周三。

Skill execution process:

Step 0: Confirm customer name "XX Trading Co., Ltd.", visit purpose "existing customer loan renewal", credit authorization "not obtained"
Step 1: No anomalies in business registration information, stable equity structure
Step 2: Judicial risk found 3 administrative penalties in the past 2 years → mark 🚩red flag
Step 3: Public opinion search found no major negative news
Step 4: Internal bank data shows credit utilization rate of 95%, deposit balance lower than 1.5 times the average monthly repayment amount → mark ⚠️yellow flag
Step 5: Credit authorization not obtained → record "skip credit inquiry"
Step 6: Compare V1-V6, no triggers, generate "pre-visit one-pager"

Output key points:

Risk signals: 1 red flag (3 administrative penalties), 1 yellow flag (liquidity tightness)
Visit recommendations: verify the rectification of administrative penalties, inquire about the reason for the high credit utilization rate, understand the repayment plan

Out of Scope

This skill does not provide credit approval decision recommendations (it only provides pre-visit risk analysis; approval decisions are handled by the credit-approval Skill)
This skill does not perform post-loan risk classification adjustments (it is only used for pre-visit background investigation)
This skill does not handle non-performing asset disposal processes (it only identifies warning signals; disposal is handled by the non-performing asset management system)
This skill does not replace on-site visits (it only provides pre-visit preparation and does not replace on-site investigation)
If the user requests the above content, clearly inform them and recommend contacting the relevant department or using the appropriate tool

About this skill

Target Role

Data Sources

Required Data

Data Desensitization Rules

Degradation Strategy

Constraints

Workflow

Step 0: Data Confirmation and Verification (read before write)

Step 1: Collection of Basic Enterprise Business Registration Information

Step 2: Judicial risk screening

Step 3: Business status and public opinion scanning

Step 4: In-bank deposit and loan data collection (if available)

Step 5: External credit reporting collection (customer authorization required)

Step 6: Veto condition self-check and report generation

Output Format

Audit Trail

Gotchas

#1: Equity penetration stops at an intermediate-level legal person

#2: Single public opinion search keyword leads to omissions

#3: Credit authorization not obtained but not explicitly recorded

#4: Visit recommendations are too general and lack specificity

Examples

Example 1: Analysis before the first visit to a new customer

Example 2: Analysis before a visit for renewal of an existing customer's loan

Out of Scope