TensorIoT helped a leading national insurance carrier switch from manual to automatic claims processing.
How do you train Amazon Textract to be able to understand and process a variety of form fields and tables?
The solution, built using Amazon Textract, processes insurance claims automatically, and has helped the client speed up their entire workflow and save money.
The client, a leading national insurance carrier based in California, processes thousands of insurance claims forms through manual data extraction. They wanted to replace that manual process with a more accurate and reliable solution that could automatically extract all relevant claim data.
While Amazon Textract makes document processing easy, the team encountered a few hurdles during this project. Non-standard form layouts have a tendency to throw Textract off. For example, a large table that was oriented horizontally, instead of vertically, made it difficult for Textract to associate the table’s labels with associated values. Additionally, Textract depends on consistency, so when form fields changed based on the users’ previous answers, the algorithms needed to be adjusted to be able to handle the variation. Finally, Textract had a hard time processing handwritten values, and the team had to come up with a solution to handle fields such as signatures.
Insurance claim forms are passed through our Amazon Textract solution, which extracts the text and data, parses and formats it, and then places it into our client’s database, using AWS Lambda triggers to seamlessly integrate the process with their document workflow. Our Amazon Textract implementation allows the client to process insurance claim forms faster and more accurately, speeding up their entire workflow process. The client now wants to explore using insights from Textract to power automatic fraud detection.