I had a conversation last week with a healthcare leader who asked me a very specific question: "Where exactly is AI working in revenue cycle right now?"
It is a fair question, and honestly, the answer is more nuanced than most vendors would have you believe. AI is delivering in some places, stalling in others, and the difference between the two has very little to do with the technology itself.
Becker’s just published a piece that captures this tension well. Their framing is that "the floor is rising" in revenue cycle, and the teams seeing the best results are investing as much in workforce preparedness as they are in the tools themselves. That last part is what caught my attention or stuck with me.
Where AI Is Actually Working (And Where It Isn't)
According to Becker’s reporting, AI has been most successful when embedded into existing workflows rather than stacked on top of them. Coding assistance, denial management, the kind of work where the AI is supporting a process that already has structure and feedback loops.
What is moving slower? Prior authorization bottlenecks, patient-facing registration tools, analytics platforms, all the areas everyone expected to be transformed by now.
Why the gap? In my opinion, it is not necessarily an AI problem. It is the difference between augmenting a workflow your team already understands versus dropping a tool into a space where the process itself is broken (or at least, not fully defined). AI can accelerate a good process, but it tends to expose a bad one.
I have seen this play out time and time again, where teams expect the tool to fix the workflow, and what actually happens is the tool reveals every assumption nobody wrote down.
The "Floor Is Rising" Reframe
The phrase Becker’s uses, "the floor is rising," is worth sitting with for a second, because it reframes the whole conversation. Most leaders are asking how high AI can take them. The more useful question, given what I have seen so far, is what AI is doing to the baseline of work everyone on the team is now expected to perform.
When a coder has AI assistance, the floor of acceptable productivity moves up. When a denial management workflow has AI triage built in, the expectation of how many denials get worked per week moves up. The ceiling is still being established, but the floor is shifting under people’s feet right now, whether they feel ready or not.
And this is where things get interesting for healthcare leaders, because rising floors are a workforce question, not a technology question.
What The Evidence Is Saying Across Regulated Industries
The pattern is not unique to healthcare. In a recent CEO study on insurance AI adoption, leaders argued that as AI touches almost every part of the value chain, human judgment, empathy and strategic decision-making become more important alongside automation, not less. The same study highlights how boards are now wrestling with model governance, bias, explainability, and how to keep AI aligned with risk appetite and fair-treatment obligations.
That is the same conversation healthcare CFOs and revenue cycle leaders are having, just with different vocabulary. Substitute "fair-treatment obligations" with "denial defensibility" and "model risk" with "coding accuracy at scale" and the questions are nearly identical.
In a related piece, "Rewriting insurance transformation by putting human experience at the center of AI", the argument is that change management cannot be delegated entirely to IT. Successful transformations require over-communication, early wins, and visible inclusion across all employee groups, including demonstrating how AI enhances roles rather than replacing them.
I would argue that in revenue cycle, this is even more acute. RCM teams have spent years being told automation was coming for their jobs. Now AI is here, the floor is rising, and the people closest to the work are being asked to adapt to tools that were often selected without their input.
What This Means For Healthcare Leaders
If you are a CFO, COO, or revenue cycle leader trying to make sense of where to invest, I want to validate something first. The fact that you are asking these questions, that you are noticing the gap between vendor promises and actual results, that is the right starting point. A lot of organizations are not even there yet.
That said, here is what I would push you to think about:
1. Audit the workflow before you audit the tool.
If prior authorization is stalling with your AI investment, the first question is not "is this the right AI?" It is "does our prior auth workflow have enough structure for any tool to accelerate it?" In a lot of cases, the answer is no, and that is not the vendor’s fault.
2. Measure workforce readiness, not just adoption.
Adoption metrics tell you who is logging in. Readiness tells you who actually understands what the tool is doing, when to trust it, and when to override it. Those are very different numbers, and the second one is what predicts whether your investment compounds or stalls.
3. Make your senior coders and denial specialists context architects, not just users.
Your most experienced RCM staff hold tacit knowledge that no AI has. The unwritten rules. Which payers play games with which CPT codes. Which denials are worth fighting and which are not. If you put AI in front of them and just ask them to "use it faster," you waste their expertise. If you ask them to teach the system (and the juniors using it) what good looks like, you build something durable.
4. Be honest about where the floor is now.
If AI-assisted coding is delivering measurable lift, the expectations for your team have changed. Pretending otherwise is not kindness, it is avoidance. Naming the new baseline out loud, and giving people the support to meet it, is what real change management looks like.
5. Resist the urge to stack tools.
Becker’s reporting on this is clear, and it matches what I see in practice. AI works when it is embedded, not stacked. Every additional tool layered on top of a workflow that nobody redesigned adds cognitive load, not capacity.
What Becomes Possible
I want to close on something hopeful, because I do think there is a lot of opportunity here that gets lost in the anxiety about AI in healthcare finance.
Revenue cycle is one of the few areas in a hospital where AI can deliver real, measurable value without touching clinical care directly. That is a gift, in a way. It gives leaders a chance to build AI muscle (governance, workforce readiness, vendor selection, measurement) in a domain where the stakes, while real, are operational rather than clinical.
The teams that use this moment well will not just see better collections or fewer denials. They will build the organizational habits, the trust between humans and tools, the muscle for honest measurement, that they will need when AI starts reaching deeper into clinical workflows. Which it will.
Honestly, the floor is rising, and that is just true. The question, in my opinion, is whether you are using that pressure to develop your people, or just expecting them to figure it out on their own.
The first path is harder up front, but it is also the only one I have seen actually work.
