AWS Entity Resolution adds record-level confidence scores for ML matching

AWS Entity Resolution has introduced record-level confidence scores for Machine Learning (ML) based matching workflows, replacing the previous group-level confidence score system. Previously, all records within a match group shared the same confidence score, regardless of individual match quality, which made it difficult to distinguish between strong and weak matches. This approach limited the ability to apply differentiated thresholds for downstream activation, often forcing customers to use a single confidence threshold across all records. The new feature provides a per-record signal of how confident the model is in each identity match, allowing for more granular control over which records are activated.

The update enables customers to apply differentiated confidence thresholds, using higher thresholds for automated merges and lower ones for additional records that still meet quality standards. This precision helps qualify more records for activation while maintaining compliance-grade transparency with audit-ready evidence for each resolved record. The existing RecordConfidenceLevel column now reflects the actual per-record score in incremental ML workflows, requiring no schema changes. This ensures backward compatibility while enhancing the accuracy of match assessments.

Customers can now start using record-level confidence scores in all AWS Regions where AWS Entity Resolution is available, expanding the feature’s accessibility. The implementation does not require changes to existing schemas, simplifying adoption for users already leveraging the service. This enhancement addresses a key limitation in the previous system, where group-level scores obscured variations in match quality within a group. The new capability allows for more targeted and efficient activation of resolved identities.

For more details, AWS provides a user guide outlining the feature’s functionality and implementation. Additional information about AWS Entity Resolution, including its broader capabilities and use cases, is available on the product page. The update reflects AWS’s ongoing efforts to improve the precision and transparency of identity resolution workflows, aligning with customer needs for more granular control over match quality.