General insurer automates claims and underwriting document intake with an AI extraction platform

Where the pressure started.

Claims and underwriting teams received large volumes of scanned documents daily — medical bills, KYC papers, damage reports, endorsement requests — mostly on email and shared drives. Junior operators keyed data into the core system by hand. Errors and rework were common, first-response times were slipping against SLAs, and there was no consistent audit trail of what a human saw versus what was entered.

Designed for the real operating environment.

• Built a document intake platform on AWS to receive documents from email, portal upload and mobile capture.
• Deployed Amazon Textract for text and table extraction, with a fine-tuned classification model to route each document to the right workflow.
• Added a validation layer where operators review only fields the model marks as low confidence, cutting review effort sharply.
• Integrated the platform with the core policy and claims system through Azure Logic Apps and secure APIs.
• Set up an evaluation dataset and monitoring dashboard so the team can track accuracy per document type as volumes grow.
• Built a governance framework covering data retention, model versioning and access controls, aligned with the insurer's regulator obligations.