Few vertical industries are as document-intensive as healthcare, whether on the insurance or provider side. That makes the healthcare industry ripe for tools that can automate insurance claims processing and other tasks.
The challenge is heightened because most of the documents in question are unstructured, consisting of text, numbers, and images that vary in nature and content. That means approaches to intelligent document processing that rely on templates to identify and extract content will be ineffective because you can’t reasonably create a template for every potential document an insurance company or healthcare provider may have to process.
Healthcare insurance providers in particular need intelligent document processing tools that can “read” a document much like a human would, extract the relevant information, and transfer it to a format suitable for input to a downstream tool, such as JSON, CSV, or XML.
Related Content: Transforming Insurance Operations with Intelligent Automation
Mountains of unstructured documents
The complexity of the situation can hardly be overstated. A large health insurance carrier can often process upwards of 3 million documents per year of various types, each one containing potentially dozens of relevant data fields. A clinical document or record could include free form text from doctors’ notes; various numbers, including ID numbers for the case, member, and provider; patient name and date of birth; date of service; various health insurance codes to depict the procedure performed and diagnosis.
On the healthcare provider side, the roles are essentially reversed, as they must deal with forms and correspondence from a multitude of 3rd party insurance carriers, each with their forms and formats.
It takes a small army of humans to process these forms, which is why health insurers and providers are investing in process automation software. Many carriers have found that robotic process automation (RPA) tools, and solutions that combine optical character recognition (OCR) with templates, can not adequately address the problem. While they work well for tasks or documents that are the same every time, they tend to hit a wall regarding all the variations inherent in unstructured documents, such as emails, patient records, and doctors’ notes.
Automating document processing for unstructured content requires a solution that incorporates technologies including natural language processing (NLP) and deep learning. These technologies are the key to the platform being able to digitize and automate highly variable documents and images without writing thousands of rules in the background.
The training component is vital. With intelligent document processing, it doesn’t take thousands of documents to train the tool. Because of intelligent automation’s unique application of transfer learning, it takes only a few hundred. From there, the IDP solution will be able to discern a discharge summary document from a medical history record document or a case ID number from a member ID number on different forms – even if the tool has never before seen that exact form or document. There’s no need to continually tweak the model each time a new type of document comes along, meaning model maintenance all but goes away.
Dramatic productivity gains with IDP
The benefit of intelligent automation tools is hard to overstate. IDP can dramatically reduce the human resources required to process mountains of paperwork – reductions around 80% are common. To learn about how it all works, download this white paper from process automation software experts at the Everest Group, “Unstructured Data Process Automation.”