Why revenue cycle is the first ‘smash hit’ application for AI

I recently attended a dinner of about 25 people who did not work in healthcare. As a talking point, I brought up the Change Healthcare ransomware attack (yes, I know, not the most casual conversation starter). That hack had impacted tens of millions of patients and cost UnitedHealth Group - Change’s owner since 2021 - close to $3 billion. It was big news, but perhaps not the front page news it deserved to be. 

And crickets. 

So few of the attendees of the dinner, including several clinicians, had heard of Change Healthcare - let alone the broader category of revenue cycle management or RCM. Change is the largest clearinghouse and medical claims processor - and it sits within an industry that’s worth more than $300 billion, conservatively. If clinical care is the brains, arms and legs of healthcare - think of the revenue cycle as its guts…  invisible to most people, until there’s a problem. And even then, the vast majority of people don’t understand what it is. But RCM deeply matters, because it's the set of processes that determine if, how and when our providers get paid, as well as whether or not medical procedures or treatments can even happen. 

“Let’s use an analogy of a McDonalds,” said Michael Gao, president of Smarter Technologies and a physician, on a phone call with me. “Imagine if the receipts on the number of burgers sold were inaccurate, then it would be hard to operationally run the store.” That observation led Gao to start his company SmarterDx, which later merged with two other companies (Access Healthcare and Thoughtful.ai) to form New Mountain Capital-backed Smarter Technologies. SmarterDx was formed to help hospitals and provider groups better analyze the care delivered, and make sure they’re capturing all the value from it. 

Smarter Technologies just announced the acquisition of Pieces Technologies to launch SmarterNotes, which connects clinical notes to quality and reimbursement. That’s another big trend in the space that we’ll get into - ambient documentation technologies are more tightly integrating with the revenue cycle. 

Per Gao and a dozen other experts I spoke to, RCM is also going through a massive transformation at this very moment, fueled by AI. New companies are forming in the space, incumbents are facing real threats for the first time in decades, and there’s an unmistakable feeling in the industry that disruption could be coming. What this could result in, according to Gao, is far fewer companies that have their own administrative cost and potentially even dollars flowing back to health systems. 

RCM is so fascinating because it’s a uniquely American industry. Per Mike Desjadon, CEO of Anomaly Insights, a company that works with health systems to manage payers, it exists because of the complexity of getting to the answer only it can facilitate: “Will this care get paid for?” 

With more than one million providers billing more than 1,000 insurance companies and government agencies across an average of 5 billion claims per year, about 850M (17%) of them get initially denied. It can cost anywhere from 3 to 6 percent of a hospital’s net revenue to manage RCM - which is essentially the cost of collecting. Most major waves of innovation in RCM over the years have focused on reducing that cost to bolster margins. But per Desjadon, RCM is entering the 4th of the major phases of change focused on addressing its own massive costs and inefficiencies.  

As he outlined them: 

Phase 1 - Better Directing Labor: From the 1990s through the mid-2000s this phase saw an explosion of business intelligence analytics tools and rule-based automation.  This gave rise to companies like MedeAnalytics, Visiquate, the Advisory Board and Optum (which acquired the Advisory Board). It also saw clearinghouses like Waystar and nThrive (now Finthrive), and Availity rapidly consolidate mid-sized vendors into the comprehensive brands they are today to bolster rapidly commoditizing core clearinghouse functionality and price. Epic also entered RCM in a massive way in this same timeframe.

Phase 2 - Offshore the labor: Wave 1 yielded a ton of spend, and some efficiencies, but didn’t produce the expected industry impact because it still required humans in the loop. As a result, the logic became outsourcing the function and offshoring the labor to reduce costs directly was the path forward. Companies that ushered in this new phase include R1, Ensemble, Conifer, Optum360, Access Healthcare and MedMetrix.  Most of these are still going strong today. 

Phase 3 - Automate the labor (rise of the bots): Outsourcing and offshoring produced massive companies, inclusive of private equity-backed players like Ensemble and R1. It also produced some efficiencies and yielded some performance, but for many reasons it proved hard to scale exponentially.  Outsourcing is right for some, but not for all. Moreover, it still requires human labor and the cost arbitrage of that labor is shrinking. As a result, Robotic Process Automation (RPA) came into vogue, particularly amongst investors. It made sense, given RCM is filled with repetitive tasks that can be automated, and bots are a heck of a lot cheaper than people. But then Olive happened. Olive’s meteoric rise underscored the huge appetite for change and the challenges associated with the phases preceding it, but its failure underscored how complex RCM really is, and left customers and investors with big questions around the execution of the technology and skepticism about technology’s ability to deliver scaled automation. 

Phase 4 - Bring in the AI and agents: The experts I spoke to believe we are now moving into the fourth phase, which could be the last one. RPA struggled in part because the technology wasn’t intelligent enough to drive itself and adapt to the ever-changing dance between providers and payers.  Simply, it can’t reason, make decisions and adapt.  RPA and early AI required too much human programming to keep up with the massive data challenge RCM is subject to.  Adaptive agents that can learn, adapt, research, and draw conclusions offer something missing in the earlier phases.

Many of the companies that got big in Phase 3 are now moving into Phase 4. The best case scenario as that occurs? 

Well, some people in our industry do believe it’s appropriate that roughly 20% of the U.S.’s GDP is spent on healthcare - with a caveat. If it’s actually spent on healthcare, meaning the delivery of care, the therapeutics, diagnostics and other interventions – versus pure administrative waste. If those dollars are getting spent on developing new drugs to save patients lives or on care delivery, or even social determinants like housing and education, the spend would be far more palatable. 

“No one wants to see $500 billion every year spent on administrative friction,” said Desjadon.

Desjadon believes the industry’s reaction to the HITECH act of 2009 offers a realistic window into what this 4th phase might look like. The HITECH act (massively) incentivized the meaningful use of electronic health records and promoted interoperability, catalyzing data standards and enabling the exchange of health information. This created a huge amount of data that never existed in healthcare. So much data, that the CAGR of health data in the US now outstrips most other industries in the world. 

This gave rise to the enormous brands we have today as they raced to consolidate every node of health data and the dollars attached to them.  The modern forms of Epic, Cerner, Athena, Optum, Change, Waystar, Availity and hundreds of others were born in this era.  They consolidated access to this explosion of data and created enormous companies. The next wave of companies, tomorrow’s winners, will be the ones to figure out 20 years later how to actually make this data useful, and actually lower the monumental cost of transacting, Desjadon feels, rather than warehousing it and profiting from the cost.  Game on.

Enter modern machine learning and agentic AI. Companies are placing big bets with the help of insatiable venture capital appetites, on the idea that there’s human labor that can be eliminated from many facets of the RCM process. In theory, that could make some of the leading revenue cycle companies a lot more efficient, and therefore more profitable. And in the best case scenario, and with the right incentives, per Gao, those dollars could flow back to the providers.

Imagine these companies take a percentage of what they help providers collect (say it’s 4%), and some portion of that goes to operations (imagine it’s 2%)... now consider that AI can make RCM companies that little bit more efficient. That’s lower costs, and in theory also higher margins.

Here’s another thought though, and Desjadon thinks this thought might just be driving the venture capital dollars flowing into this space… 

What if this is a classic “faster horses” argument, and the real opportunity is eliminating the revenue cycle as we know it?  

“Why not?” he says.  “The whole point of RCM is to figure out if a provider will get paid, and if so, how much and when. It's so convoluted and expensive because that’s a hard answer to get. But what if it wasn’t? If we knew that answer at the time of every encounter, how much of the process would we really need?  That is the power of today’s technology.” (He’s not alone in making that argument. We’ll explore that theme throughout the piece). 

In any case, most RCM entrepreneurs seem to be attracted to it because of the opportunity to use AI for manual tasks. 

“A lot can be automated already,” said Sam Schwager, CEO of SuperDial, a company that is tackling one of the bottlenecks – the phone calls required for benefits verification and claims status updates. “It’s a game with a complex set of rules, but that’s also what makes it the first smash hit application for AI.” 

Schwager believes there will be massive efficiencies in the next few years, and that RCM will be the first industry in healthcare with case studies that show the practical value.

Beyond AI, and also because of AI, I asked a range of experts in the space about the most important changes that are underfoot. All the experts I spoke to for the piece agreed that the space is highly fragmented, which is where the potential lies in aggregating the various solutions, as well as opportunity for new technology. 

“I’ve seen far more emphasis on the front end, meaning to make the patient more of a stakeholder” said Seth Cohen, president of Cedar, in the patient-facing payments space. For those unfamiliar, there’s three stages to RCM, and these terms are commonly used in the industry: Front end, mid cycle, and back end. Payers and providers would have a different set of processes for each.  

Front-end has historically received very little attention, but now increasingly is center stage as providers are looking to get ahead of any potential billing-related problems. The category includes but is not limited to the following set of processes: 

“Providers are finding ways to get everything done upfront before the patient even comes in,” Cohen said. “This is ultimately to reduce risk to the provider.” As he explained, the last thing a provider wants is for a patient to get seen, and then to learn that the provider is actually out-of-network and a large bill they can’t afford is on the way. For a long time, RCM really didn’t touch on the patient experience much at all. Patients were on the receiving end when they received a bill. Now, with much larger balances to pay, they are increasingly viewed by the industry as a stakeholder. What that means in practice is setting clear expectations ahead of the visit, and giving them consumer tools they’re used to, like electronic billing options and digital navigation. 

There’s also efforts across a lot of provider groups to do the prior authorization work upfront, as well as to determine medical necessity. With patients on high deductible plans facing high costs, providers are also moving to share cost estimates with patients ahead of time. Cost transparency is still nowhere near perfect, noted Cohen, who’s been working on it throughout his career, including while working at Castlight Health. As with almost everything in healthcare, it’s far more challenging than it seems on paper.

On the one hand, it’s surprising that a lot of employed physicians aren’t deep on the revenue cycle market. On the other hand, it is not. “There’s no visibility and it’s very opaque for providers, especially ones employed by a larger system – what you do and what you get paid,” said Ash Zenooz, a physician and longtime health technology operator. “It’s only where you leave or you run your own practice where it becomes relevant and important because it’s part of the value chain.” 

Many of the operators I spoke to in the space noted that the problem is in the opacity related to this space. Most physicians and those providing care didn’t get into medicine to do battle with insurance companies. Physicians shouldn’t have to learn about the healthcare revenue cycle or engage it in their everyday work and lives, but unfortunately for now they increasingly are required to. Zenooz noted it’s what is needed for survival, especially for independent practitioners. She also said that RCM is increasingly the “operating system” for healthcare, not the EHR, because everything revolves around the payment (not just the clinical notes or the patient communications). 

It’s not just about the models. The company that will be the big winner in the RCM space has access to clean, structured data that can yield the best insights and mitigate hallucinations. PE-backed Ensemble Health Partners, for instance, has been building its own data lake, and claims to have access to 25 billion transactions. Importantly, it can map those to outcomes. For non-clinical uses, “the data access (across the industry) is quite good,” said Cohen, and there’s more APIs coming out on the clinical side. 

That said, Zar from Provana argues that access to APIs needs to be talked about far more than just AI. With AI alone, and insufficient access to APIs, companies will need to create AI models that are constantly learning what the ever-changing payer rules are. With APIs, there’s theoretically far less need for that because a payer is giving visibility into their rules. But many payers today still don’t provide much visibility into their rules, and even subcontract them out to payment integrity vendors that may be incentivized to deny claims. 

There seems to be a running theory in the RCM space that there could actually be a disruptor to the clearing house, and there do seem to be some viable contenders in the market today inclusive of private-equity backed, potentially disruptive players like Availity. In theory, there could be a large aggregator like Epic or Athena that has enough data to go directly to the payer and cut out the clearinghouse. The reason that hasn’t happened yet is because of all the R&D and time - it takes real resources to go to hundreds of payers and build out the integrations. But most of the folks I spoke to believe this could be on the roadmap in the next three or four years. 

What success would look like, per the RCM operators I spoke with, is less RCM. “The real opportunity is to eliminate the work,” said Zar. “That means digitizing the processes and interactions between a payer and provider to ensure that about 90 percent of the time there’s a payment without any work.” Zar acknowledges that there will be complex edge-cases about 10 percent of the time, but he believes that even that work can get optimized. 

He doesn’t believe that any provider in the future should have 20 people employed just to get prior authorizations for medical claims. AI agents will be doing that in a matter of months, not years. “I’m working myself out of a job everyday,” he said. 

That also dovetails with another optimistic view within the industry that the promise of AI and other technologies is that it could bring payers and providers closer together, versus exacerbate existing rifts. That may be the only way to cut hundreds of millions of dollars in inefficiency and waste. 

One of the challenges with inpatient care is that there may not be a conversation between a patient and provider that could be recorded by an ambient scribe, and ultimately be a good signal for medical decision making and care provided. 

What’s needed, according to Gao, is a way to determine that in scenarios where there’s no voice-based discussion. An “evidence trail” is needed. He said that the ultimate goal is to code in a way that’s justifiable for the complexity of the care delivery. 

The optimistic thesis is that all this new technology makes it easier for health systems to do the pre-work and potentially even bill upfront, which reduces the fractional cost of the administrative side of health care. All that reduces the amount of cents needed to collect the $1.

“By decreasing the cost of the administrative side,” said Gao. “My hope is that we can maximize the fraction of healthcare dollars that flow to patient care.”