Sales Quality & Fraud Prediction

Problem & Business Impact

Fraudulent or poor quality sales (falsified documents, inconsistent profiles) generate financial losses, regulatory risks and early churn.

Measured impact:

• -50% detected activation fraud
• +35% overall sales quality
• +40% regulatory compliance
• Savings: 2-3% of sales revenue

Data & Key Features

Risk signals

• Documents: quality, consistency, falsification
• Customer profile: age/income/product consistency
• Suspicious patterns: mass activations, single address
• Agent: quality history, fraud rate
• Timing: off-hours activation, abnormal speed

Calculated scores

• Fraud score (0-100)
• Sales quality score
• Compliance risk
• Recommended action (validate/block/investigate)

Models & Methods

1. Fraud Detection: Classification

   • Accuracy: 91%
   • Fraud recall: 87%
   • False positives: <5%

2. Anomaly Detection: Suspicious patterns

   • Isolation Forest
   • Outlier detection

KPIs & Performance Targets

| Metric | Target | Current | |--------|--------|---------| | Detected fraud | -50% | -52% | | Sales quality | +35% | +37% | | Compliance | +40% | +42% | | False positives | <5% | 4.2% |