RPA to Gen AI: How AI in revenue cycle management is evolving

From RPA to generative AI

Nearly three-quarters of healthcare organizations automate at least some parts of revenue cycle management, and about half of those organizations are using AI to automate, according to a recent survey from AKASA. The most common automation technology among health systems and hospitals represented in the survey was robotic process automation.

RPA isn't really AI, Tabatabaie explained. "It's essentially the art of screen scraping, so if a person is putting keystrokes on the left field on the top of the screen, the system learned what they key in," he said.

But more and more in revenue cycle, you are better off using generative AI if the right solutions exist because there's a big gap between the clinical information and the information that lives in revenue cycle.

"So, you can tell the system to grab pieces of data and put in in this other form," he continued. "Not really AI, but it's very broadly used for tasks that are very repeatable and tasks that are very transactional."

Where healthcare organizations are getting more technical is layering AI with RPA solutions, as 30% of healthcare financial leaders told AKASA. Organizations are also now leveraging natural language processing.

"NPL has also been around for a while, but it is essentially teaching a machine what a language means," Tabatabaie said. "You can let it know that this date is an age versus the date for termination of a contract, for example. The machine will learn to understand that, and it understands negation. Again, it's been around but it has obvious limits on how much it can learn."

NPL solutions are only as good as what you teach them, Tabatabaie emphasized.

This is why healthcare providers are getting more excited about advanced AI capabilities, such as generative AI. Made popular by ChatGPT, revenue cycle management has now set its sights on self-learning capabilities or deep learning within AI solutions, and for a good reason.

"It's the future of AI," Tabatabaie said. "With generative AI, you can ask the system to do things or answer questions, and it goes and figures out how to do it based on a lot of knowledge that it has gathered and deciphered. Then, it generates answers from what it has learned elsewhere, not what human beings are teaching it."

Healthcare organizations are using all three of these technologies, and there's a place for them all, Tabatabaie said.

"But more and more in revenue cycle, you are better off using generative AI if the right solutions exist because there's a big gap between the clinical information and the information that lives in revenue cycle," he said. "Generative AI can look and learn from the clinical side and decipher what the right thing to do is and fill in the blanks with a lot more context."

The state of generative AI adoption

With about a quarter of healthcare organizations still relying on manual processes for revenue cycle management, AI is not considered a staple in the healthcare revenue cycle. However, most healthcare financial leaders expect to continue adopting automated solutions, with 90% of revenue cycle leaders expecting generative AI to play a bigger -- and beneficial -- role in medical coding operations.

Tabatabaie said healthcare is at an inflection point with AI in revenue cycle management.

"The early adopters have suffered through dealing with vendors that don't have all the T's crossed and the I's dotted," he elaborated. "Now, the second phase is coming to the market, and there are several driving factors, both technical and non-technical."

From a non-technical standpoint, healthcare organizations are simply out of talent at a time when the administrative burden is only increasing. Medical coders and other revenue cycle staff may also be retiring over the next couple of years.

Additionally, offshoring tasks like medical coding may not produce the quality level organizations require as denial rates tick up. Administrative tasks are also contributing to physician burnout, driving more organizations to automate tasks to lighten the workload.

Technically, machines using AI can reveal insights from data that revenue cycle staff wouldn't necessarily be able to glean from their datasets, according to Tabatabaie.

"Machines can look for the right codes for the right case, but we can also learn a lot more about the patient and the physician, like length of stay, results and outcomes," he explained. "All that data is available to go deep and get more insights out of it."

This is just the start for generative AI, too. Healthcare organizations may start by using generative AI for simple coding, but when they discover denial patterns, for example, they can start to prevent denials in the first place.

"Also, the machine sees when you submit a case for reimbursement to Aetna versus Blue Cross Blue Shield, for example," Tabatabaie said. "So, you go from simple coding in the revenue cycle to denials prevention, then you may start to identify whether or not a patient has care gaps if billing or registration information is missing. Machines may be able to gather information from schedules to alert you that there may be a problem or a patient needs outreach."

Down the line, generative AI solutions may even be able to identify patients traveling to sunnier locales and at higher risk of skin cancer, Tabatabaie used as an example.

"In a non-invasive way, we may be able to look out for a patient's welfare with the amazing amount of information we can get," he said. "The skin cancer patient booked a flight to Florida, so we can contact them about skincare products."

This use of AI may sound far off, but it is a real possibility.

"Just like we started to use the internet for very mundane tasks, now you can't live without it, and we even expect websites and smartphones to keep our credit card numbers," Tabatabaie stated. "The evolution has begun."

A word to the wise

AI is improving revenue cycle management and has the potential to continue down the optimization path. However, Tabatabaie said AI isn't magic; it will not solve all of the revenue cycle's problems.

"You give the machines bad data, and either they can't do their job right, or they flat out make mistakes," he explained. "The idea of buying a solution and all my troubles going away is a fallacy."

It's about a commitment to data flowing with high-quality, rapid rates and measuring the output to make sure it justifies the reasons why you bought it.

Healthcare organizations must have their technical ducks in a row, so to speak. "Make sure the information technology side of the house has bought into the project and is keeping in step with what's needed to produce a good outcome," Tabatabaie said.

Historically, getting buy-in for AI in revenue cycle management has been a challenge, though. Many organizations prioritize innovation for patient care or clinical workflows. After all, the mission of a hospital is to keep patients and their communities healthy. This has led to implementation and deployment missteps in the revenue cycle.

"It's about a commitment to data flowing with high-quality, rapid rates and measuring the output to make sure it justifies the reasons why you bought it," Tabatabaie said.

Tabatabaie recommended keeping "a human in the loop," especially when applying AI to a task like coding. Medical coders are key to submitting clean claims even though AI and automation are making it easier to translate clinical speak to the language of medical and billing codes.

AI should be making revenue cycle staff work smarter, rather than taking their jobs away.

"We have so much shortage of talent on an ongoing basis that if you remove the mundane tasks, people can take advantage of that and pay attention to more difficult cases, more difficult tasks," Tabatabaie said. "So, AI companies should have this component called human-in-the-loop so humans can help the machine."

Jacqueline LaPointe is a graduate of Brandeis University and King's College London. She has been writing about healthcare finance and revenue cycle management since 2016.