Scoring Engine Overview

The Scoring Engine is SEON's core decisioning system, designed to empower businesses with advanced fraud prevention capabilities. By leveraging comprehensive transaction analysis, it allows companies to assess risk effectively through customizable rule sets and machine learning models. The system provides an adaptable framework to identify and mitigate threats in real-time, ensuring a proactive stance against fraud.

SEON enriches incoming transaction data by supplementing it with multiple data points derived from the API request. Each data point undergoes analysis to detect potential risk factors, and the engine then calculates a fraud score. This score determines the likelihood of fraudulent activity, and transactions are categorized into one of three possible states based on the score and rule application:

• APPROVE: Transactions are determined to be low-risk and allowed to proceed without additional intervention.
• REVIEW: Transactions require further manual examination to confirm authenticity before approval.
• DECLINE: Transactions are identified as high-risk and automatically blocked to prevent potential fraud.

Example Rules

• "No online profiles found" (+7 Fraud Score): Adds 7 points if a provided phone number cannot be linked to any social media or digital platform. Lack of such associations raises suspicion.
• "Email is not deliverable" (+4 Fraud Score): Assigns 4 points when an email address is undeliverable, indicating potential invalid data entry.
• "User disabled cookies" (DECLINE): Blocks the transaction if cookies are disabled, as fraudsters often use this method to avoid tracking.

The final decision within SEON involves a combination of scoring rules, list management, and machine learning evaluations. The blacklist and whitelist enforce strict blocking or approval policies by setting fraud scores to 100 and 0, respectively. Custom rules offer greater flexibility, allowing nuanced adjustments to the score based on contextual data. Blackbox scores, while separate, can be integrated into rules for enhanced accuracy by evaluating complex data patterns. The process ensures full transparency, with rule evaluations and results included in API responses and the event details page.

Rules are central to SEON's Scoring Engine, defining how data points influence both fraud scores and transaction outcomes. SEON provides a structured approach to rule management through three distinct categories:

Default Rules

Default rules are pre-configured by SEON's fraud prevention specialists, designed to cover commonly observed fraud patterns. While users can toggle them on or off, they cannot be modified or deleted.

Custom Rules

Custom rules provide businesses with the flexibility to create tailored transaction analysis strategies. Users can:

• Modify fraud scores based on specific conditions.
• Adjust transaction states (APPROVE, REVIEW, DECLINE).
• Add data points to blacklists, whitelists, or custom lists.

Machine Learning Rules

Machine learning rules are auto-generated from historical transaction data and can be enabled manually or based on accuracy metrics. These rules help businesses adopt adaptive strategies that evolve with their unique risk profiles.

SEON integrates two specialized machine learning models to enhance decisioning accuracy and automation:

Whitebox Machine Learning

Whitebox ML generates transparent, human-readable rules based on past transactions. It provides accuracy metrics and continuously improves through data labeled via the Label API, allowing ongoing refinement of the classification logic.

Blackbox Machine Learning

The Blackbox model calculates a separate fraud probability score ranging from 0 to 100 - called Blackbox score - independent of the standard fraud score. It identifies complex correlations and risk patterns that may not be evident through rule-based scoring alone, serving as a secondary layer of defense.

Effective list management is a crucial component of SEON's fraud prevention strategy, helping control decisioning outcomes. SEON supports three core list types:

Blacklist

Blacklists automatically assign a DECLINE state to transactions containing blacklisted data points, such as compromised IP addresses, device hashes, or user IDs. This is suitable for known fraudulent entities.

Whitelist

Whitelists automatically assign an APPROVE state for trusted data points, such as verified customer email addresses or employee IDs, minimizing friction for legitimate users.

Custom Lists

Custom lists enable the monitoring of specific data points without immediately impacting the fraud score, while also allowing their evaluations (such as checking if an item is on a custom list) to be used in rules to influence scores. They can also be used to group customers based on certain values and data fields.

SEON's rule categories facilitate efficient rule organization and scoring management. They allow businesses to:

• Group rules by specific fraud patterns or use cases.
• Assign category-level scores and states to control outcomes more precisely.

By implementing rule categories, businesses can ensure a structured approach to fraud prevention while maintaining transparency and efficiency in transaction management. This structured, layered decisioning process ensures both accuracy and flexibility in combating evolving fraud tactics.