Imagine that you’re the CEO of a large healthcare provider, and you’re thinking about what to do with AI. You’ve heard about some of the fascinating results out of various research labs about how AI can equal or exceed human physicians in diagnosing cancer, retinal diseases, and even Covid-19. You salivate a bit at the dollars you might bring in from funding organizations and rich donors who want to be associated with these sexy developments.

But then you snap out of that dream and focus on the real value of AI for your hospitals. First of all, you know that what works in the lab doesn’t always work at the clinical bedside. Secondly, you remind yourself that the FDA has to approve these AI treatments, and that can take years. The last realization to wipe out your dream is that it’s unlikely any of these AI diagnostic capabilities will reduce your costs; you’ll probably still need the same number human radiologists, ophthalmologists, and so forth.

What you really need, it finally occurs to you, is AI to reduce administrative costs. You remember a conference you attended at a lovely resort at which Professor David Cutler, a renowned health economist from Harvard, was speaking. He said that administrative costs take a quarter to a third of total health-care spending in the U.S. He advanced several ideas for lowering costs related to the revenue cycle, prior authorization, and data clearinghouses. The CEO wasn’t sure, but he believed that the professor had mentioned that AI might be able to help with some of these approaches.

This CEO’s sudden realization about administratively-oriented AI has been successfully accomplished at Baylor Scott & White Health, a large 52-hospital academic medical system that is the largest not-for-profit provider in Texas. Sarah Knodel, the Senior Vice President of Revenue Cycle there, may not be the CEO, but she oversees many administrative functions and has 2,500 people reporting to her.

Knodel told me that the BSWHealth system has three major focus areas relative to the revenue cycle:

• Reduce their cost to collect
• Optimize net revenue
• Improve the patient financial experience

She feels that AI and automation can help with all of them—sometimes simultaneously.

Knodel and BSWHealth have been working on one approach to these areas for eight years now. With a focus on improving pricing transparency for its patients, BSWH implemented an automated, machine learning-based price estimation tool from Waystar, a healthcare technology vendor. The tool generates estimates of patients’ out-of-pocket costs before they receive care. That may seem ordinary in many industries, but it is very unusual in the healthcare sector to be able to accurately estimate costs in advance. Prior to implementing the tool, creating the estimates was a very manual process of combining disparate information from numerous systems at BSWHealth. It took a revenue cycle employee 5 to 7 minutes to produce one estimate, with limited accuracy. Now, however, 70% of the estimates are calculated without any human touch. The system automatically retrieves real-time eligibility and benefit data from the patient’s insurer and combines this with charges and contracted rates to create an estimate of out-of-pocket costs unique to a specific patient. The technology gathers and learns from insurance claims to improve the accuracy of estimates over time.

No outside organizations evaluate hospitals on the patient financial experience, but BSWHealth has received positive feedback since implementing the tool. Payment options are discussed in advance of care, leading to 60-100% improvements in point-of-service collections across various clinics and hospital departments. Physicians are also happy with estimates being provided in advance, because it leads to fewer cancellations of procedures on the day of service. Five years ago Knodel recommended that the estimate system be made available in an online self-service format so price shoppers could obtain their own estimates when evaluating where to receive care. Over the last year the U.S. government mandated such advance estimates for care, but BSWHealth was well out in front of the requirement (and many hospitals are not yet compliant with it).

BSWHealth is also leveraging intelligent technology for “claim statusing” in the business office’s insurance collections department to automate the process of checking the status of outstanding insurance claims. In the past, a human collector would have to log into multiple payer websites or call them. Now, a robotic process automation (RPA) and screen-scraping technology mimics the user signing into the payer website. As the RPA system gets the claim status from payers, the data is integrated into the workflow of the collector such that it never hits the collector’s work queue if it’s accepted and scheduled to be paid. Conversely, accounts that are denied and require immediate action are accelerated for review. The statusing RPA results in an exception-based workflow where only accounts truly requiring human intervention are brought forward for review by a collector.

Sarah Knodel said that her organization is undertaking many projects of this type and using machine learning or RPA in almost all revenue cycle departments. In areas like utilization review, new technology reads medical record documentation in real time and predicts whether a patient should be in inpatient or observation status, ensuring compliance with regulatory and payer requirements. As a result of that effort, BSWHealth has reduced FTEs in the utilization review department by over 20% while reducing payer denials by the same percentage. Looking forward, Knodel’s goal is to use these technologies to develop more collaborative and innovative partnerships with payers. She hopes to eliminate the time-consuming and inefficient back-and-forth process of treatment authorizations and appeals in favor of something more automated and efficient.

The Waystar Perspective on a More Intelligent Revenue Cycle

To understand what’s happening in the broader industry on AI and RPA for revenue cycle and administrative cost reduction, I spoke with Matt Hawkins, the CEO of Waystar. That company is also a good example of how transaction processors can create new offerings from their data exhaust. He said that Waystar processes 2.5 billion transactions a year in healthcare billing and collections, all on a single data platform, for about 40% of U.S. patients. That data allows the company to leverage machine learning, other forms of AI, and RPA to help take costs out of the system and apply them to better patient care.

Nobody ever accused the U.S. healthcare system of being simple, and that’s reflected in the data about it. Waystar works with 18,000 healthcare provider organizations and a large number of billing and collections firms. Waystar provides a clearinghouse for billing and collections data, but in order to analyze it and apply AI they have to retrieve, match, standardize, federate, and transform the data with rule engines. After that, they can apply all sorts of algorithms, like the one to predict how much a procedure will cost at will cost at one of the large health systems using their technology. Their AI and RPA platform is called “Hubble.” It doesn’t allow a view into other galaxies, but it does allow a view into the future of your insurance claims.

Some of Waystar’s offerings provide “propensity modeling” of one type or another—predicting the probability that a claim will be judged by the payer as accurate and will pay it, predicting whether a patient can pay their bill or not, etc. That all involves machine learning.

Waystar uses RPA to automate things like claims statusing, claims denial appeals, and other administrative processes that involve checking data from one system or another and communicating about it. They are having robots ping or call payer firms, and increasingly the payer firms are having robots handle the response. Pretty soon we humans will mostly be watching robots talk to another.

Perhaps seeing the machine learning projects in use at Baylor Scott & White Health and the AI and RPA applications that Waystar offers with revenue cycle might be tempting to our hypothetical CEO who’s trying to figure out what to do with AI technology. It seems unlikely that a big donor would sponsor a big initiative in that space, but then again the rev cycle work would probably more than pay for itself in short order.