Amazon Bedrock and HealthLake automate processing of medical insurance claims

AWS has published a detailed guide to building an automated AI pipeline for processing medical insurance claims, based on Amazon Bedrock and AWS HealthLake.

Two tools, one process

Bedrock Data Automation handles intelligent data extraction from insurance claim forms. It analyzes unstructured documents, including scans, PDFs and handwritten forms, and converts them into machine-readable structured fields.

Amazon Bedrock AgentCore hosts and runs an AI agent that takes the extracted data, validates it against defined rules and transforms it into FHIR resources (Fast Healthcare Interoperable Resources), the international standard for exchanging medical data. The finished resources are written directly to AWS HealthLake, a fully managed cloud store for medical data.

The two services work in sequence as a single pipeline: Data Automation handles document recognition, while AgentCore handles validation and transformation logic.

How the process works

The pipeline runs through several sequential stages that together replace manual processing:

• Form upload: the insurance claim enters the system (CMS-1500, UB-04 or other standard formats)
• Data extraction: Bedrock Data Automation parses the document and pulls out fields: patient data, ICD-10 diagnosis codes, CPT procedure codes, payout amounts and dates of service
• Validation: AgentCore runs an AI agent with a set of business rules that checks code correctness, date consistency and the presence of required fields
• Transformation: the agent converts the verified data into FHIR-compliant resources: Patient, Claim, Coverage, Organization
• Write to HealthLake: the finished resources are saved to the store, where they are available for analytics, auditing and further routing

Automatic checks at every stage catch errors before anything is written. Manual intervention remains only for non-standard cases.

Why FHIR in insurance

FHIR is an open HL7 standard adopted by hospitals, insurers and regulators worldwide. In the US, it has been mandatory in the CMS (Centers for Medicare & Medicaid Services) system since 2021. Converting data to FHIR at the start of the pipeline eliminates repeated conversions when passing information to government registries, hospital systems or BI platforms.

"You will learn how to combine these services into an end-to-end workflow that reduces manual processing while maintaining accuracy through automated checks," — AWS ML

Blog.

AWS HealthLake supports FHIR R4, built-in resource search, analytics through Amazon Comprehend Medical and export to S3. For insurers with large claim volumes, this provides a standardized output format without developing custom converters.

What this means

AWS is steadily building a vertically specialized stack of AI tools for healthcare. The combination of Bedrock Data Automation, AgentCore and HealthLake turns the manual processing of insurance forms, one of the most labor-intensive processes in health insurance, into a managed, automated pipeline on fully managed services. Teams without MLOps experience do not need to train their own document recognition models or build FHIR integrations from scratch.