Scaling Prior Authorization Automation with IDP

The current prior authorization processes were some of the most resource-intensive workflows for this North American healthcare system. They serve more than 10 million people each year, with 33 hospitals across the United States. As the system grew through clinic acquisitions, so did the volume and complexity of managing Personal Health Information (PHI) and administrative documents. Capacity to process and manage these documents needed to scale, with prior authorization (PA) processing at the top of the list.

Multi-page packets arrived daily, blending typed forms, supporting documentation, reference material, and handwritten physician notes that carried the real clinical nuance behind a request. As volumes grew, each packet added to prior authorization processing delays.

The existing intelligent document processing (IDP) configuration treated every packet the same way, pushing all pages through ABBYY Vantage regardless of content complexity. Vantage handled structured typed text well but was far less accurate on handwriting. That meant handwritten documents fell into manual work queues, and required full re-keying.

At the same time, the system was consuming units on every page, including unnecessary reference material. As a result, processing full packets took longer and generated exceptions that were escalated to humans for review, often on documents that did not include the critical fields required to move the PA forward.

Ashling set out to build a solution that would improve the quality and speed of PA processing, especially for handwritten notes, while keeping platform usage and costs under control.

We began by reframing PA as a split-path problem, treating each page of a packet based on its content type. Handwritten notes were routed to Chorus for specialized extraction, while typed pages flowed through ABBYY Vantage. In testing, Chorus read handwriting at over 95 percent accuracy, while ABBYY delivered approximately 86 percent. That 9 percent difference was key to ensuring reliability in the extraction of critical health data fields.

This approach depended on one key capability: splitting the packet at intake.

To make this vision work, we first needed to reliably separate typed and handwritten pages at intake. We turned to ABBYY Vantage’s new “Splitter” skill, an OCR-powered AI introduced in 2023 that uses structural and visual cues to understand where one document ends and another begins.

There was a catch: Splitter was brand new, largely untested, and light on training data.

Our team built one of the first custom-trained Splitter models for healthcare PA and improved it through iterative human feedback. Over time, the model learned to distinguish handwritten notes from supporting typed documents with increasing precision, unlocking the full potential of the split-path architecture.

A Blue Prism wrapper then brought everything together into a single end-to-end flow. It orchestrates the entire process from ingestion through Splitter and routing to Chorus or ABBYY Vantage. Analysts have clear visibility into each job’s status, can see exactly where it sits in the flow, and can troubleshoot exceptions from a managed queue.

The PA solution became a flagship example within a broader IDP transformation. At the program level, the health system scaled from 6 million to 40 million documents processed per year while improving cost control, traceability, and audit-ability. Exception volumes dropped, accuracy improved, and analysts spent more time on meaningful edge cases instead of manual re-keying work.

The client now has a repeatable pattern for structured and unstructured document intake, and a solution architecture that keeps accuracy high and cost low as volumes grow. Components of this design can now be reused and applied to other clinical and back-office operations, serving as a Blueprint for scale.

Next, the health system is exploring agentic solutions to close the loop. Trained on policies and guidelines, an AI agent can use the extracted context to recommend approval or denial decisions for each PA claim. By moving from manual review to guided decisioning, patients receive faster, more consistent answers, and staff spend less time on manual review. The result is a PA experience that feels smoother for patients and measurably lighter for the teams who support them.

Technology Used