EXL Medical IDP cuts claims review from 100+ minutes

The EXL Medical IDP solution addresses a longstanding operational problem in insurance and healthcare: medical records are long, unstructured, and full of specialized clinical terminology that demands domain expertise to interpret. EXL, a data analytics and AI provider serving Fortune 500 organizations for over 25 years with more than 50,000 professionals globally, built the solution on AWS to keep model development and production inference under one roof. The pipeline combines Xtrakto.AI, a template-agnostic document processing application that uses computer vision, NLP, and agentic AI workflows, with the EXL Insurance LLM, a domain-specific model fine-tuned on nine years of insurance claims operations data comprising over 13,500 records. Amazon SageMaker AI hosts the training and inference, while Amazon Bedrock provides access to general-purpose foundation models for broader language tasks.

A key design choice was fine-tuning rather than prompting general-purpose models. EXL argues that prompting alone cannot consistently handle insurance-specific tasks such as medical record annotation, economic and non-economic damages summarization, and negotiation guidance. The fine-tuning used Parameter-Efficient Fine-Tuning with Low-Rank Adaptation on SageMaker AI, which adapts the model without modifying all parameters of the base model. In internal benchmarking by EXL, the fine-tuned model performed well on tagging, summarization, question-answering, and reasoning, assessed with automated metrics and blind review by three insurance subject-matter experts. The pipeline also includes responsible-AI controls built into the deployment, such as content-filtering and grounding checks, source-level traceability, and de-identification procedures that align with HIPAA requirements.

A large healthcare payer deployed the solution to replace fragmented manual workflows, connecting systems including EPIC, CarePort, and Predictal through a single API-led orchestration. The results reported by EXL include reduced manual effort, accelerated member outreach, and increased clinical bandwidth without additional headcount. The outcome likely hinges on whether the confidence-based routing and human-in-the-loop validation continue to maintain data integrity as case volumes grow, since the solution is designed for augmentation rather than replacement of human expertise. For claims adjusters, underwriters, and care coordinators, the promise is less time on data retrieval and more time on judgment, though the reported gains come from EXL's internal benchmarking and a single payer deployment.