Confidence Scoring & Explainability

Introduction & Philosophy

Our APIs provide detailed explanations for every resolution attempt, helping you understand both the confidence level and reasoning behind each result. This explainability system is actively developed in collaboration with our customers to provide flexible yet standardized insights into our AI agent's decision-making process.

How Confidence Scores Work

Confidence scores range from 0-100, representing how certain our agent is in the output results. This score is generated through a robust process that combines multiple scoring mechanisms behind the scenes, which we continually update and improve to enhance accuracy.

Factors Array Overview

The factors array breaks down the reasoning behind each confidence score into two categories:

Strengths: Evidence that supports the match or resolution
Limitations: Weaknesses or uncertainties that lower confidence

Each factor includes:

• Standardized codes for programmatic processing
• Impact classification (strength vs limitation)
• Natural language descriptions explaining the reasoning
• Severity levels for limitations (low, medium, high, critical)

Actions Array - Improving Your Results

Actions are suggested improvements that appear when confidence scores are lower due to unclear or insufficient input data. These provide actionable steps to get better results in future queries.

Common scenarios where actions help:

• Generic company names with multiple matches
• Missing location data when multiple entities exist
• Ambiguous legal structures (Inc. vs LLC with same name)

Real-World Example

Scenario: User searches for "Smith Construction" without location

Why low confidence: Even a human couldn't confidently distinguish between:

• Smith Construction Inc. (manufacturing company)
• Smith Construction LLC (residential contractor)

Suggested action: "Provide location (city/state) or legal structure to distinguish between multiple entities"

Schema Structure

{
  "factors": [
    {
      "type": "limitation" | "strength",
      "code": "string",
      "description": "string", 
      "impact": "string",
      "severity": "low" | "medium" | "high" | "critical"
    }
  ],
  "actions": [
    {
      "code": "string",
      "description": "Detailed description of the recommended action"
    }
  ]
}

Common Factor Codes

Limitation Codes (Lower Confidence)

• no_location_provided: User did not provide location in query
• multiple_potential_entities: Multiple entities match search criteria
• name_too_generic: Name provided is too broad or common
• location_mismatch: Provided location doesn't match entity information
• insufficient_data: Not enough information found during research
• unsupported_entity_type: Entity type not yet supported (government, person, etc.)
• no_legal_entity_confirmed: No official legal entity was confirmed

Strength Codes (Higher Confidence)

• name_exact_match: Provided name exactly matches legal entity name
• location_confirmed: Provided location matches registered location
• jurisdiction_match: Entity jurisdiction matches query jurisdiction
• legal_entity_confirmed: Valid legal entity confirmed through official sources
• multiple_sources_corroboration: Multiple authoritative sources confirmed details
• registration_number_match: Provided registration number matched entity

API-Specific Variations

Factor and action codes vary slightly between Entity Resolution and Domain Intelligence APIs, reflecting the different data inputs and resolution processes.

Learn more:

• Entity Resolution API for complete code references
• Domain Intelligence API for complete code references

Using Explainability Data

For Developers

• Parse factor codes programmatically to understand result quality
• Use action suggestions to improve subsequent queries
• Monitor limitation patterns to optimize your data inputs

For Business Users

• Review natural language descriptions to understand match confidence
• Follow action recommendations to get more accurate results
• Use factors to validate whether results meet your quality standards

Next Steps:

1. Try it: Entity Resolution API - See confidence scoring in action
2. Learn more: Entities and Candidates - Understanding when entities vs candidates are returned
3. Advanced: Domain Intelligence API - Confidence scoring for domain analysis