EXL has built an AI pipeline to reduce the manual work of reviewing medical records for insurance claims and life underwriting. Adjusters can spend more than 100 minutes on a case that contains hundreds of pages, according to the company, and must connect diagnoses, treatments, tests and clinical notes across many document types.
The system first uses Xtrakto.AI to split, classify and extract information from records without fixed templates. A specialist EXL Insurance language model then provides summaries, natural-language queries and structured outputs. The model was fine-tuned for medical and insurance terminology, including diagnosis and procedure codes, while general-purpose models available through Amazon Bedrock handle broader language tasks.
The AWS architecture uses Textract and Lambda for document preparation, SageMaker endpoints for specialist-model inference, Step Functions for orchestration and S3 for source documents and results. Access is controlled through AWS identity services, and the workflow runs within one AWS region. EXL says outputs preserve traceability so reviewers can connect generated answers to supporting information.
This is a vendor case study, not an independent clinical or claims-accuracy evaluation. The published description does not justify replacing professional judgment, especially where records are ambiguous or a decision affects coverage. Its practical role is to organize and surface evidence faster while adjusters and underwriters remain accountable for the final determination.