GPT based entity matching solution to improve operation efficiency by over 400%

Operational improvement using GenAI
Customer Challenge

A healthcare customer had been using fuzzy matching and manual effort to match records to ensure correctness and completeness of data. The fuzzy matching was automatically matching only 15-20% of the records and the rest of the 80% of the records were being manually matched. Typical data volume that was processed by the customer is about 100,000 – 150,000 records per month, hence the manual effort was slowing down the process performance.

Solution Implemented

Evaluated multiple GenAI and Non-GenAI solutions to validate applicability and built a matching solution that leverages the power of LLMs (GPT3). Solution allows users to review the matches and the omissions. The integrated solution allows continuous monitor of models for performance and refresh the model embeddings on a regular basis.

Outcome Delivered
  • Improved match accuracy by over 400%.
  • Reduced manual processing time from weeks to hours.
  • Allows end-users the ability to approve / reject matches thus allowing for a feedback loop that will further improve model performance over time.