Engineering partner for SaaS & platform companies.
Book a Strategy Call →Have a smaller task? Submit a quick request →Copyright 2025© Peregrine-it.com | All Rights Reserved
How we built an InsurTech claims automation platform processing 45K+ claims per year — with document extraction, multi-carrier adjudication rules, and a full audit trail, achieving 60% faster processing and HIPAA compliance.
InsurTech Claims · HIPAA Compliant
A regional health insurer processing 45K+ medical claims per year was drowning in manual adjudication. Claims adjusters opened PDFs and EOBs one by one, typed data into spreadsheets, and checked carrier rules against printed policy documents. A single claim could take 20+ minutes. Backlogs stretched to weeks, and providers complained about delayed reimbursements. Errors were common — wrong procedure codes or duplicate submissions slipped through.
They needed an automation platform that could extract structured data from uploaded documents (PDFs, scanned images) using OCR and ML, apply multi-carrier adjudication rules to determine eligibility and payment amounts, maintain a complete audit trail for compliance and disputes, and integrate with HL7 FHIR for interoperability with provider systems. HIPAA compliance was mandatory — all PHI had to be encrypted, access-logged, and retained according to policy.
Adjusters manually read each claim form and supporting documents. Data entry was error-prone — wrong CPT codes, missing modifiers, duplicate submissions. A single adjuster could process only 20–30 claims per day.
Claims arrived as PDFs, scanned images, and sometimes faxes. Extracting procedure codes, dates, amounts, and patient identifiers required manual review. OCR tools existed but produced noisy output with no validation.
Different carriers and plans had different reimbursement rules, exclusions, and prior-auth requirements. Rules changed frequently. There was no centralized engine to evaluate claims against policy logic.
Auditors and dispute resolution required a complete record: who viewed what, when decisions were made, and what data was used. Legacy systems had partial logs and no way to reconstruct claim lifecycle.
We built a claims automation platform with a Next.js portal for adjusters and a Node.js backend that orchestrates document ingestion, extraction, adjudication, and payment. AWS Textract extracts text and tables from PDFs and images; we run validation and entity resolution against our schema. A rules engine evaluates each claim against carrier-specific configs. HL7 FHIR integration enables exchange with provider systems. Every action — view, edit, approve — is logged with user, timestamp, and payload for audit.
Document extraction was the hardest part. Textract returns raw text and table cells — we had to build logic to map fields to our schema (patient ID, provider NPI, procedure codes, dates, amounts). We added confidence thresholds: below 85% confidence, we flag for human review. That reduced auto-processing errors by 40% while still automating the majority of straightforward claims.
Audited claim types and workflows. Documented carrier rules and exclusion logic. Evaluated Textract vs. alternatives. Designed claim schema and audit log model. Defined HIPAA controls and encryption strategy.
Integrated AWS Textract with custom field mapping. Built rules engine with carrier-specific configs. Implemented validation and confidence scoring. Added human-in-the-loop queue for low-confidence extractions.
Built Next.js claims portal with review queue and approval workflow. Implemented HL7 FHIR endpoints for provider integration. Added full audit logging and encryption. Conducted HIPAA risk assessment.
Piloted with 5K claims. Tuned extraction and rules based on error analysis. Scaled to 45K+ claims/year. Trained adjusters and documented runbooks. Achieved 60% faster processing and passed compliance review.
Claims automation sits at the intersection of document intelligence and business rules. Textract gives you raw extraction — the real work is mapping that to your schema, validating against known codes (CPT, ICD-10), and handling the long tail of document formats. We built a validation layer that catches common errors (wrong date format, invalid procedure code) before claims reach the rules engine. Low-confidence extractions go to a human review queue; we don't auto-approve those.
Multi-carrier rules are inherently complex. We modeled rules as configurable expressions — eligibility checks, reimbursement formulas, exclusion lists — stored in versioned JSON. When a carrier updates policy, we add a new version with an effective date. The engine evaluates claims against the correct version for the service date. That kept rule maintenance manageable as we added carriers.
HIPAA compliance meant encrypting PHI at rest and in transit, logging every access, and implementing role-based controls. We worked with the client's compliance team from day one — audit requirements shaped the schema and logging design. The full audit trail turned out to be a competitive advantage: when providers disputed claims, the client could show exactly what data was used and when decisions were made.
We help InsurTech companies build document automation, adjudication engines, and compliant claims systems. Let's talk about your architecture.