AI is dominating healthcare conversations, promising to revolutionize operations — including revenue cycle management. But the hard truth, according to some RCM experts, is that most AI-driven RCM solutions will fail. The reason: AI without deep operational expertise doesn’t work in the real world.
To understand where AI is driving measurable change in RCM and where it’s headed, Becker’s Healthcare recently spoke with Billy Parrish, chief product officer, and Andrew Ray, chief transformation officer at Ensemble, a company that manages revenue cycle operations for 27 health systems nationwide.
Question: AI is being touted as a game changer in RCM, but issues like payer-provider friction persist. What are the biggest barriers to solving this challenge?
Billy Parrish: Solving the complexities of the healthcare revenue cycle with AI at scale is particularly challenging for a few reasons. One is high variability in payer policies. Coverage guidelines and payment rules differ between payers and plans, with hundreds of changes issued monthly. This variability makes it difficult for AI systems to standardize processes and predict outcomes accurately across different payers.
Additionally, payers and providers use diverse systems and formats for data collection and storage, which are often not designed to communicate with each other. This lack of interoperability means that data from different sources can be inconsistent and incomplete, making it challenging for AI algorithms to process and analyze information effectively. Without standardized data, AI solutions struggle to deliver accurate and reliable results.
Q: Where is the greatest potential for improvement and impact in RCM?
BP: Reducing friction starts with understanding patterns by analyzing historical data. We’ve been using various forms of AI for a decade to decrease manual, non-value-added drudgery in the revenue cycle. Using rich data to drive emerging capabilities like generative AI will have a tremendous impact on RCM efficiency and quality.
Andrew Ray: Historically, RCM relied on human expertise. Something got denied and a person had to figure out what the issue was and how to resolve it. Now, technology will learn, iterate, improve and streamline the entire process. But even the best AI is limited by the data it’s trained on, so having well-understood, normalized data — and having it at scale — is critical to the success of AI in healthcare and in RCM.
Using agentic AI — on its own or in combination with generative AI — to perform non-value-added RCM tasks is the next step. The technological capability is improving so fast that the process and performance improvements will be enormous.
Q: Ensemble emphasizes combining AI with deep RCM expertise. Can you share more about Ensemble’s approach and how you ensure that solutions deliver meaningful, lasting impact?
AR: We rely on a “human-in-the-loop” approach to ensure the ethical, effective application of all technology. Operational expertise and oversight is foundational for our AI, which we’re using to improve performance and optimize complex processes.
BP: Ensemble, due to its scale and experience, has amassed more than 10 years of normalized, standardized, well-understood data on RCM practices that drive award-winning results. This includes data showing what actions operators took and what worked or didn’t work to achieve desired outcomes. We mine this data to detect patterns and reverse engineer root causes of payment delays and denials to create upstream interventions and automations. Ensemble’s technologists wake up thinking about emerging AI capabilities, asking, “How can this revolutionize healthcare?”
AR: We recently launched our clinical AI engine to generate accurate, context-sensitive clinical insights, including clinical denial appeal letters across a range of patient conditions. Through an iterative process involving clinicians, we’ve carefully calibrated our AI engine to create appeal letters that are now comparable to human-generated letters. When clinicians review human- and AI-generated appeals side by side, they have no preference for one over the other. That’s a huge step toward replicating complex clinical reasoning.
We’re using this capability to help clinicians focus time on reviewing and refining clinical responses instead of creating initial drafts which allows them to operate efficiently at the top of their license. And not only are these clinical summaries generated quickly, but they are incredibly rich. AI can review an exponentially greater amount of information than a clinician ever could, like a patient’s entire history and external materials, to create a more detailed response.
Q: What should healthcare leaders consider when evaluating AI-driven RCM tools?
BP: In evaluating an AI solution, it is important to look at the decisioning components driving the algorithms. These decisioning components include the data used to train models and the people and processes involved. How are answers achieved? What is the reasoning and decision chain? Are clinical staff involved? How successful has this been in making decisions? What evidence or credentials support a solution’s ability to handle complex decisions?
AR: With AI, efficacy is essential. Ensure the company has the expertise and data to train and maintain an effective AI solution. If they aren’t driving successful RCM outcomes without AI, don’t expect them to succeed with AI.
