Understanding and Improving Confidence Scores

What is the Confidence Score?

Important: The "confidence" field is actually a predicted error percentage, not a traditional confidence score. This naming can be counter-intuitive since lower values indicate higher confidence in the prediction.

In hindsight, this field would be more accurately named predictedError since that's what it actually represents:

• Lower confidence value = Higher actual confidence
• Higher confidence value = Lower actual confidence

Understanding the Values:

• A confidence score of 0.05 (5%) = High confidence - the model expects ~5% error
• A confidence score of 0.30 (30%) = Low confidence - the model expects ~30% error

Think of it as: "How much error do we predict in this estimate?" rather than "How confident are we?"

Important Note: A higher confidence score (indicating predicted error) doesn't necessarily mean the estimate itself is inaccurate. It simply means the model has less certainty due to factors like property uniqueness, limited comparable data or not enough details in the request. The estimate may still be very reasonable and useful for decision-making.

How to Reduce the Confidence Score (Improve Accuracy)

The accuracy and confidence of property estimates depend heavily on the completeness and quality of data provided in your request. Here are key strategies to achieve lower confidence scores (which means higher accuracy):

1. Provide Comprehensive Property Details

Include as many property attributes as possible in your estimate request.

2. Ensure Data Accuracy

Verify that all provided information is current and accurate:

• Double-check square footage measurements
• Confirm the year built from official records
• Validate property tax amounts from recent assessments
• Ensure address details are complete and correctly formatted

3. Include Location-Specific Information

Precise location data helps the AI model find better comparables:

• Provide complete address with proper formatting
• Specify the correct boardId for the MLS region (Required when using an API key that has access to multiple boards)

4. When to Expect Higher Confidence Scores (Lower Accuracy)

Some property types and situations naturally result in higher confidence scores (indicating lower accuracy):

Unique or Rare Properties:

• Custom-built homes with unusual features
• Properties on very large or irregularly shaped lots
• Historic homes with significant modifications
• Properties in areas with few recent sales

Insufficient Market Data:

• New developments with limited comparable sales
• Rural properties with sparse comparable data
• Properties in rapidly changing markets

Incomplete Information:

• Missing key property details
• Outdated or inaccurate property information
• Vague or non-specific information about property extras

Troubleshooting High Confidence Scores (Low Accuracy)

If you're providing comprehensive data but still seeing high confidence scores (indicating lower accuracy), consider:

1. Review your data completeness - Ensure all available property details are included
2. Verify data accuracy - Cross-check information against official records
3. Consider property uniqueness - Rare or unique properties naturally have higher confidence scores (lower accuracy)
4. Contact support - Our team can investigate specific cases and provide guidance

When reaching out for support, please provide:

• The complete estimate request payload
• The returned confidence score (remember: lower = better)
• Any specific concerns about the property or market area

Best Practices Summary

✅ Do:

• Include all available property details in your request
• Use accurate, up-to-date information
• Provide complete address and location data
• Consider property uniqueness when interpreting confidence scores

❌ Avoid:

• Submitting requests with minimal property details
• Using outdated or estimated property information
• Expecting high confidence for truly unique properties
• Ignoring the relationship between data quality and confidence

Updated on: 16/06/2025