Verisoul Scoring System

Verisoul’s fraud detection system uses a comprehensive scoring mechanism to help you identify and prevent fraudulent activities. This page provides an overview of how our scoring system works.

Score Range and Interpretation

All Verisoul scores follow a consistent scale:

  • Range: 0 to 1 (or 0% to 100%)
  • Interpretation:
    • 0 = Highly trusted / legitimate user
    • 1 = Highly suspicious / likely fraudulent

This consistent scale makes it easy to understand and compare different risk signals across your user base.

{
    "account_score": 0.15,  // 15% - Low risk
    "bot": 0.87,  // 87% - High risk
    "multiple_accounts": 0.08,  // 8% - Low risk
}

Decision and Account Score

At the highest level, Verisoul provides two key outputs to simplify your decision-making process:

Account Score

The account_score is the comprehensive risk assessment that combines multiple signals into a single value. It represents the overall likelihood that an interaction is fraudulent.

This score is a combination of:

  • Multi-accounting signals
  • Risk signals (account and session)
  • Bot detection

Decision

The decision field provides a string representation that maps directly from the account_score. This gives you an actionable recommendation based on the combined risk assessment.

The decision values are:

  • "Real" - Low risk, trusted user
  • "Suspicious" - Medium risk, may require additional verification
  • "Fake" - High risk, likely fraudulent
{
  "account_score": 0.82,  // 82% - High risk
  "decision": "Fake"
}

The threshold for the decision can be configured in the Verisoul’s system to match your risk tolerance.

Score Categories

Verisoul provides several specialized risk scores that focus on different aspects of fraud:

Account Risk Score

Evaluates the overall risk associated with an account based on persistent identity signals and historical behavior patterns. Learn more in Account vs Session.

Session Risk Score

Assesses the risk of the current interaction based on device information, behavioral biometrics, and contextual data. Learn more in Account vs Session.

Multi-Accounting Score

Detects when a single user is creating multiple accounts, often to abuse promotions or evade restrictions. Learn more in Multi-Accounting.

Bot Score

Identifies automated scripts and bot activity attempting to interact with your application. Learn more in Bot.

Using Scores Effectively

Threshold Setting

While Verisoul provides the decision field as a recommendation, you can also set your own thresholds based on your risk tolerance:

  • Conservative (lower false positives): Block only very high scores (e.g., > 0.9)
  • Aggressive (lower false negatives): Block moderate to high scores (e.g., > 0.7)

Contextual Application

Consider applying different thresholds based on the context:

  • Higher scrutiny for high-value transactions
  • Lower friction for returning users with good history
  • Stricter controls during promotional periods

Progressive Security

Implement a tiered approach to security based on risk scores:

  1. Low scores: Frictionless experience
  2. Medium scores: Additional verification steps
  3. High scores: Block or require manual review

The Probabilistic Nature of Fraud Detection

It’s important to understand that fraud is not deterministic but constantly evolving. Fraudsters continuously adapt their tactics to bypass detection systems. This is why Verisoul returns machine learning model scores rather than binary yes/no decisions.

These scores represent predictions about the likelihood of fraudulent activity based on patterns observed in the data. The probabilistic approach allows for:

  • Nuanced risk assessment that captures the degree of suspicion
  • Flexibility in how you apply security measures based on your risk tolerance
  • Continuous improvement as new fraud patterns emerge

The Verisoul Data Advantage

Verisoul benefits from a powerful data flywheel effect that enhances model accuracy:

  • Cross-industry insights: By analyzing fraud patterns across multiple industries, Verisoul can identify emerging threats before they become widespread
  • Broad customer base: Data from diverse customers creates a rich training dataset that improves model performance
  • Continuous learning: Each new fraud attempt detected contributes to improving the system for all customers

This data advantage allows Verisoul to train models with the highest possible accuracy, staying ahead of evolving fraud tactics while minimizing false positives that could impact legitimate users.