An Affordable Artificial Intelligence Solution for Intelligent Document Processing of Faxed Documents

Despite widespread adoption of electronic health records (EHRs), health systems remain heavily dependent on faxed documents for critical patient information. At New York University Langone Health, this represents nearly 20 million document-pages per year - laboratory results, consult notes, imaging prescriptions, refill requests, and prior authorizations - each requiring manual review and indexing. These workflows are time consuming, involve multiple staff touchpoints, can be prone to error, and may create delays for patients awaiting follow-up care. To provide the highest quality of care to patients and to augment staff experience, the authors developed and deployed an Intelligent Document Processing (IDP) solution leveraging existing enterprise technologies for document management, robotic process automation, data classification and extraction, and EHR-integrated indexing. This solution identifies electronically faxed documents, extracts patient and provider information, matches the EHR record, sorts the documents into clinical or administrative queues, and assigns a document type for indexing. To ensure patient safety, documents that cannot be confidently processed are routed to an exceptions folder for manual review. The IDP solution was deployed and monitored at one high-volume multispecialty practice from August to October 2025. In this time, the system processed approximately 20,000 document-pages, representing 13,700 faxes or scans. Of these, 8500 (62%) were successfully classified to one of the predefined in-scope clinical and administrative document types that the system was trained to recognize (e.g., laboratory results, pathology and radiology reports, procedure notes such as colonoscopy or endoscopy, medication- and insurance-related authorizations, and consult or therapy reports); based on the classification, they were then routed to the appropriate work queue for indexing. The remaining 38% required manual review - 32% were identified as being outside the target set of document types, and 6% were flagged as exceptions (e.g., multiple patients in one fax, document longer than 20 pages). The cost to operate was approximately 1.5 U.S. cents per page during the pilot, significantly less expensive than competitive industry offers of approximately 15 U.S. cents per page. Implementation required not only technical integration, but also operational redesign. Key hurdles included applying existing technologies to a single orchestrated solution, managing the unclassified documents workload, aligning document type taxonomies between systems, handling provider name variation, and training clinical staff. Change management was paramount, as individual practices had developed varied and entrenched fax workflows that required reengineering and preproduction dress rehearsals prior to go-live. This experience demonstrates the potential for an artificial intelligence (AI)-enabled IDP solution to meaningfully reduce administrative burden, improve timeliness and accuracy of document indexing, and unlock structured data from scanned pages. Never before had these practices been able to quantify and route faxed documents automatically. Although challenges remain in scaling across diverse workflows, this case illustrates how health systems can pragmatically deploy AI using existing infrastructure to improve efficiency, reduce staff burden, and support better care delivery.