Insurance Tech Startup Cuts Fiduciary Compliance Review from Manual Analyst Hours to Minutes with Claude on AWS Bedrock
At a glance
Self-funded health plans carry ERISA fiduciary duty to catch improper claims payments, but manual review leaves billing errors undetected. An InsurTech startup, which has analyzed $16B+ in claims, had no engineering team to build the AI layer its market demanded, capping how fast it could scale oversight.
NMD built a Claude-powered, prompt-based reporting system on AWS Bedrock that queries claims data using Comprehend Medical and OpenSearch for grounded, hallucination-resistant results. Staff and plan administrators now ask natural-language questions instead of waiting on manual review.
The company now runs a defensible, auditable AI query layer for a business where a wrong answer is a fiduciary liability event, not an inconvenience. Three funded AWS engagements, POC to production, position it to extend AI claims intelligence across its full plan book and industry-wide.
Industry
Use Case
Compliance & Governance, Claims Processing, Intelligent Document Processing
Solution implemented
- Claude on AWS Bedrock powers prompt-based, natural-language querying of pre-pay and post-pay claims data.
- Amazon Comprehend Medical extracts medical entities and terminology for accurate, context-aware compliance analysis.
- Amazon OpenSearch delivers semantic search across claims and plan documents for grounded query results.
- MCP-based integration connects directly to the client's Redshift Serverless data infrastructure for live queries.
- Terraform, GitHub Actions CI/CD, and AWS Cognito RBAC support production-grade LLMOps and secure access.
- Query results cache to S3 via Glue and Athena for downstream Power BI visualization.
The value equation
- Turns fiduciary compliance from manual review into auditable, natural-language claims queries at scale.
- Reduces reliance on scarce claims analysts for compliance verification and payment-integrity reporting.
- Establishes a HIPAA-compliant AI architecture built to withstand ERISA-level accuracy and audit demands.
- Converts a single POC into a repeatable, three-phase funded delivery model with AWS.
- Positions the company to extend AI-verified oversight across its full self-funded plan portfolio.
Company Snapshot
An Insurance Tech startup providing payment integrity and fiduciary compliance for self-funded health plans, having analyzed more than $16 billion in claims and identifying an average of $82.37 per member per year in improper payments.
Location
United States
Customer Situation
This InsurTech startup protects self-funded health plan assets by reviewing claims for improper payments and fiduciary compliance, work governed by ERISA and the Consolidated Appropriations Act. The company had analyzed more than $16 billion in claims and identified $82.37 per member per year in improper payments, but compliance verification still depended on manual analyst review of plan documents and a roughly 170-field claims schema.
The company wanted a prompt-based system to extract compliance rules from plan documents and answer natural-language questions about claims data, but had no internal engineering resources to build it. Without a scalable AI layer, it risked being outpaced by better-resourced competitors in a market where speed and accuracy protect plan sponsors from fiduciary liability.
NMD Solution
NMD reviewed and found that the client’s manual review process could not scale to the query volume its self-funded plan clients needed, and that generic AI querying risked hallucinated compliance findings, an unacceptable risk under ERISA. The solution required a medically grounded architecture: AWS Bedrock and Claude for prompt-based reasoning, Amazon Comprehend Medical to extract clinical entities, and Amazon OpenSearch for semantic retrieval, all deployed inside a HIPAA-compliant, audit-logged AWS environment integrated with the client’s own Redshift Serverless data layer.
What We Delivered
The company’s staff, brokers, and self-funded plan administrators query pre-pay and post-pay claims data in natural language and receive medically grounded, auditable answers in place of manual analyst review. The production system integrates directly with the client’s Redshift Serverless infrastructure through an MCP-based connection, extracts compliance rules from summary plan documents, and returns structured results to Power BI through an S3, Glue, and Athena data cache. The team is now tuning the semantic-context layer to extend accuracy and scale as query volume grows from hundreds toward the tens of thousands originally projected.
Ready to Transform Your AI Software Business?
See how New Math Data can transform your software business with AWS-powered innovation.