DOCUPIPE
Solutions
Resources
Pricing
Bardavon Health automates referral intake with DocuPipe
Bardavon Health Innovations is a workers' compensation care coordination company that manages treatment for injured workers across the United States. With document extraction powered by DocuPipe, Bardavon standardizes referral intake packages, pulling structured data from medical records, prescriptions, and insurance documents to support case management.
Challenge
When an injured worker is referred to Bardavon, the process starts with a document package. Insurance adjusters, employers, physicians, and nurse case managers all send documentation - and every referral looks different.
A single intake package can exceed 50 pages: emails from adjusters, impairment ratings, handwritten prescriptions, faxed therapy orders, state injury forms, clinical notes, and follow-up instructions. Some arrive as phone photos of paper. Others are barely legible faxes.
Bardavon needs to extract diagnosis codes, treatment prescriptions, and employer details fast enough to begin coordinating care.
"Coordinating care for injured workers involves loads of documentation. We receive documents from insurance companies, employers, physicians, therapists, utilization review, and many others. To deliver our services efficiently, we must be experts at extracting key data points and making quick decisions," says Daniel Hochee, Engineering Manager at Bardavon Health Innovations.
The company had tried AWS Textract, but couldn't exceed 75% success rates given the document variety and quality - meaning 25% of documents still required manual processing.
Solution
Bardavon needed a system that could handle the full range of document quality and formats in workers' comp referrals: handwritten prescriptions, faxed therapy orders, state forms that differ by jurisdiction, and clinical notes with dense medical terminology.
DocuPipe's proof-of-concept proved it could handle the variety. Bardavon built a Referral Intake schema that standardizes incoming documents into structured fields - ICD-10 codes, demographics, and key case details - regardless of source format.
Staff use DocuPipe's Review feature to validate flagged results with bounding boxes and confidence scores. They see exactly where each data point was extracted and only review items flagged with lower confidence.
Within weeks, extraction success rates exceeded 90% - reducing documents requiring manual intervention from 25% to under 7%, a 4X improvement.
Results
4X reduction in manual intervention
With AWS Textract, 25% of documents required manual processing. With DocuPipe, that dropped to under 7% - a 4X reduction in documents needing human intervention.
20X faster claims processing
DocuPipe processes referral documents 20X faster than manual workflows. Documents that previously required staff to read, interpret, and transcribe now extract automatically - staff only verify flagged low confidence items.
14X volume growth, same workflow
In four months, monthly page volume increased 14X (from ~600 to over 9,000 pages) without changing the DocuPipe extraction setup. New referral sources and document types are handled by the same schema.
Review-based quality assurance
In workers' comp, data accuracy is critical. A misread diagnosis code can delay treatment or create compliance issues. DocuPipe's Review feature shows bounding boxes for each extracted value. Staff verify only flagged items, confirm or correct, and move on.
Schema tuning in plain English
Bardavon improves extraction by editing schema instructions in natural language. Changes take effect immediately. When new document types appear, the schema adapts with a text update.
With intake bottlenecks removed, care coordinators spend less time on paperwork and more time connecting injured workers with the right therapists.
We tried implementing automated data extraction using AWS Textract, but given the wide variety and quality of documents, we could attain no better than 75% success rates. DocuPipe has come to the rescue. It's developer-friendly and integrates seamlessly into our workflow, and keeps improving with simple tweaks to our instructions.
Read more
Ready to get started?
Create an account and start extracting data from documents - no contracts required. Or, contact us for enterprise pricing.