Risk Scoring

VeriPlus calculates risk scores (0-100) for applicants, wallets, and transactions using machine learning and rule-based models.

What is Risk Scoring?

Risk scoring quantifies the likelihood that a customer or transaction is fraudulent, involved in money laundering, or otherwise high-risk.

Benefits:

• Automate decisions - Accept low-risk, reject high-risk, review medium
• Prioritise resources - Focus compliance team on high-risk cases
• Reduce friction - Fast-track low-risk customers
• Audit trail - Defensible risk-based approach for regulators

Risk Score Range

Applicant Risk Score

Scoring Factors

VeriPlus calculates applicant risk from:

Document Fraud Score

AI analyses documents for manipulation indicators:

Fraud Score:

• 0-30: Authentic
• 31-60: Uncertain, manual review
• 61-100: Likely forged, reject

Biometric Confidence

Face match score (0-100):

Liveness Detection:

• Pass: -5 risk
• Fail (deepfake detected): +50 risk

AML Risk

Screening results affect risk:

Country Risk

Based on FATF ratings and corruption index:

Calculation Example

Base Risk Score: 50

Factors:
+ Document fraud score: 12 (-12 points, good document)
+ Face match score: 92 (-5 points, strong match)
+ Liveness: Pass (-5 points)
+ AML: PEP Tier 2 match (+30 points)
+ Country: United Kingdom (0 points)
+ Historical: First-time user (0 points)

Final Risk Score: 50 - 12 - 5 - 5 + 30 + 0 + 0 = 58

Risk Level: HIGH
Recommendation: Enhanced due diligence

KYT Wallet Risk Score

Scoring Factors

Cryptocurrency wallet risk (0-100):

Category Scores

Each category scored 0-100:

Example Wallet:

Sanctions: 0 (no sanctions)
Darknet Markets: 45 (traced funds)
Ransomware: 8 (distant connection)
Scam: 12 (low exposure)
Stolen Funds: 6 (minimal)
Mixer: 38 (mixed funds detected)
Gambling: 23 (some gambling)
Exchange: 65 (high exchange interaction)

Weighted Average: 45 × 0.30 + 8 × 0.25 + 12 × 0.20 + 38 × 0.15 + 23 × 0.05 - 65 × 0.10
= 13.5 + 2 + 2.4 + 5.7 + 1.15 - 6.5
= 18.25

After sanctions check (0, no sanctions):
Final Risk Score: 18

Risk Level: LOW

Sanctions Override: Any sanctions match = automatic 100 risk score

Transaction Depth

Risk calculated at different trace depths:

Example:

Depth 0 (wallet itself): Risk = 10 (clean)
Depth 1 (direct sources): Risk = 35 (1 source is mixer)
Depth 2 (2 hops back): Risk = 62 (traced to darknet)
Depth 3 (3 hops back): Risk = 78 (ultimate source is ransomware)

Recommendation:
- Depth 1-2 for standard checks
- Depth 3 for high-value customers

Transaction Risk Score

Real-Time Scoring

For each transaction, VeriPlus assesses:

Velocity Rules

Monitor transaction volume:

Time Window: 24 hours
Transactions: 15
Total Value: $45,000
Average: $3,000
Max: $15,000

Risk Assessment:
- High transaction count (+20)
- Large total value (+25)
- Large individual transaction (+15)

Transaction Risk: 60 (HIGH)

Structuring Detection

Detect attempts to evade reporting:

Example:

Transactions in 48 hours:
$2,900, $2,800, $2,950, $2,850

Pattern: Multiple transactions just below $3,000 threshold

Risk Flag: Structuring
Risk Score: +40

Risk-Based Workflows

Low-Risk Flow

Risk Score: 0-30

Actions:
1. Auto-approve
2. No manual review
3. No additional checks
4. Normal monitoring

Medium-Risk Flow

Risk Score: 31-50

Actions:
1. Standard verification
2. Basic AML screening
3. Approve with monitoring
4. Quarterly re-screening

High-Risk Flow

Risk Score: 51-70

Actions:
1. Enhanced verification (liveness)
2. Comprehensive AML screening
3. Manual review by compliance team
4. Source of funds verification
5. Monthly re-screening

Critical-Risk Flow

Risk Score: 71-100

Actions:
1. Reject automatically OR
2. Senior compliance review
3. In-depth investigation
4. Law enforcement notification (if sanctions)
5. Continuous monitoring if approved

Custom Risk Rules

Create organisation-specific rules:

Rule Examples

Rule 1: High-Value Customers

if (transactionValue > 10000) {
  riskScore += 20;
  requireManualReview = true;
}

Rule 2: New Account Activity

if (accountAge < 30 && transactionCount > 10) {
  riskScore += 30;
  requireEnhancedDueDiligence = true;
}

Rule 3: Country + Amount

if (countryRisk === 'HIGH' && transactionValue > 5000) {
  riskScore += 40;
  rejectAutomatic = true;
}

Machine Learning Model

VeriPlus uses ML to improve risk scoring:

Training Data

Model trained on:

• 10 million+ verification records
• Known fraud cases
• Regulatory enforcement actions
• Industry fraud patterns

Features

ML considers 150+ features:

• Document metadata
• User behaviour patterns
• Device fingerprints
• Time of day, day of week
• IP geolocation
• Email/phone providers
• Typing patterns

Model Performance

Continuous Improvement: Model retrained monthly with new data

Risk Score API

Get Applicant Risk

GET /api/v3/applicants/:id
 
Response:
{
  "applicantId": "app_123",
  "riskScore": 58,
  "riskLevel": "HIGH",
  "riskFactors": [
    {
      "factor": "AML_PEP_MATCH",
      "impact": +30,
      "description": "PEP Tier 2 match found"
    },
    {
      "factor": "DOCUMENT_QUALITY",
      "impact": -12,
      "description": "High-quality authentic document"
    }
  ]
}

Get Wallet Risk

GET /api/v3/kyt/checks/:id
 
Response:
{
  "checkId": "kyt_abc123",
  "riskScore": 72,
  "riskLevel": "HIGH",
  "categoryScores": {
    "darknetMarkets": 45,
    "ransomware": 8,
    "mixer": 38
  }
}

Audit and Compliance

Risk Score Explainability

Every risk score includes:

• Component breakdown - Which factors contributed
• Impact values - How much each factor changed the score
• Recommendations - Suggested actions
• Timestamp - When score was calculated

Example:

{
  "riskScore": 65,
  "components": [
    { "name": "Document Fraud", "score": 15, "weight": 0.35 },
    { "name": "Biometric Match", "score": 85, "weight": 0.20 },
    { "name": "AML Screening", "score": 40, "weight": 0.25 }
  ],
  "rationale": "High risk due to PEP match and medium country risk",
  "recommendation": "Enhanced due diligence required"
}

Defensibility

Risk scores are:

• Auditable - Full calculation breakdown
• Reproducible - Same inputs → same score
• Explainable - Human-readable reasoning
• Compliant - Meets regulatory standards

Best Practices

1. Set Clear Thresholds: Define accept/review/reject scores for your risk appetite
2. Review Regularly: Audit risk model performance monthly
3. Adjust for Business: Tune thresholds based on false positive/negative rates
4. Document Exceptions: Record why high-risk customers were approved
5. Monitor Drift: Track if risk score distribution changes over time
6. Combine with Manual Review: Don't rely solely on automated scoring
7. Update Rules: Adjust risk rules as fraud patterns evolve

Next Steps