1. Introduction

1.1 Blacklist Decision Rule

The Blacklist score is derived from telecom behavioral data, including SMS, voice, and call center interactions. It is recommended that clients evaluate Blacklist with the MAW for a better outcome.

1.1.1 True Blacklist Definition

A user is classified as a True Blacklist when:

• L5 risk signals (from MAW) are observed in:
• SMS data
• Automated voice (IVR/robot call) data

These signals are considered strong indicators of high-risk behavior and should be treated as confirmed blacklist events.

1.1.2 Call Center Data Exclusion

Call center (agent) records (based from MAW) should not be directly included in blacklist count calculations because they contain a mixture of operational activities, including:

• Marketing calls
• Customer service interactions
• Payment reminders
• Collections activities

Furthermore, collection activities may represent significantly different risk profiles:

Therefore, call center records should be used as supplementary risk indicators rather than direct blacklist evidence.

---

2. Risk Assessment Framework

2.1 New Customer Scenario (Acquisition)

New customer applications typically represent the majority of business volume and require a balance between risk control and approval rate optimization.

2.1.1 Positive Indicators (Approval Enhancement)

The following characteristics may indicate lower risk:

SMS / Voice Behavior

• Multiple L1 or L3 records (from MAW)
• No L4 or L5 records (from MAW)

Call Center Behavior

• Presence of L5 call center records (from MAW) with actual connected call duration
• Stable communication behavior demonstrated based from MAW through:
  • Consistent answer rate
  • Meaningful call duration

Negative Indicators (Approval Reduction)

The following characteristics may indicate elevated risk:

Call Center Behavior (based from MAW)

• Multiple dialing attempts within a recent period (e.g., 7 days)
• No successful call connection
• Zero call duration

SMS / Voice Behavior

• Multiple L4 hits (from MAW)
• Any meaningful occurrence of L5 signals (from MAW)

Multi-Lending Behavior (based from MAW)

• CID hit count exceeds business threshold
• Example: More than 10 CID matches Such patterns may indicate excessive borrowing activity or credit stress.

---

2.2 Existing Customer Scenario (Credit Line Increase)

For existing customers, historical behavior should be incorporated into the risk assessment process.

2.2.1 Positive Indicators

Communication Behavior

• L5 call center records with normal connected call duration (from MAW)

Borrowing Activity based from MAW

• Moderate CID hit volume
• Example: 3–5 CID matches

Repayment Stability

• No obvious loan discontinuation risk
• CID activity remains within a reasonable range

2.2.2 Fraud Indicators

Special attention should be paid when:

• CID hit count equals 0
• CID hit count equals 1

Extremely low borrowing activity may indicate:

• Synthetic identity risk
• Insufficient behavioral history
• Potential fraud attempts Additional verification is recommended before making a final decision.

---

3. Parameter Configuration Guidelines

3.1 Observation Window

The optimal observation period should be aligned with the customer's business model and product characteristics.

3.1.1 New Customer Applications

3.1.2 Existing Customers

Observation windows should be adjusted according to:

• Product tenure
• Credit cycle
• Customer lifecycle stage

---

3.2 Threshold Calibration

Thresholds should not be fixed across all customers. Recommended calibration factors include:

• Dial attempt frequency
• Unanswered call rate
• Call connection rate
• CID hit count
• L4/L5 occurrence frequency

Thresholds should be periodically reviewed and optimized based on portfolio performance and business objectives.

---

4. Recommended Decision Strategy

Instead of relying on a single blacklist indicator, it is recommended to combine:

• Blacklist signals (L4/L5) from MAW
• Communication behavior
• Multi-lending indicators (CID)
• Recency and frequency metrics
• Customer lifecycle information

A multi-factor risk assessment framework generally provides better predictive performance and more stable approval strategies than using blacklist counts alone.