Most health systems’ first brush with AI in the revenue cycle looks the same: a bot that checks claim status, a tool that flags a missing modifier, an assistant that drafts an appeal letter. These deployments are worth doing.
Nearly two-thirds of healthcare organizations now use AI or automation somewhere in the revenue cycle, and that share is expected to keep climbing as systems look to offset staffing shortages and improve cash flow, according to HFMA. McKinsey estimates that enabling the revenue cycle with AI could reduce cost to collect by 30 to 60 percent, according to its recent analysis. But speeding up an existing task and redesigning how work gets done are two different ambitions, and only one of them creates a lasting advantage.
From Faster Tasks to a Different Operating Model
Traditional machine learning has already delivered real value in the revenue cycle: predicting denials, flagging missing documentation, routing claims. Agentic AI is a different kind of tool. Beyond executing a predefined rule, it can reason across an entire workflow, decide what to do next, and act, with a person stepping in only where judgment is required.
McKinsey has called this the first credible path to a “touchless” revenue cycle. We’d put it a bit more cautiously: Agentic AI is a credible step toward a lower-touch revenue cycle, not a touchless one. Complex claims, denials and appeals still run through payer-specific rules, exceptions and judgment calls that no model fully captures yet.
That distinction changes the question health systems should be asking. Instead of “where can AI assist a person?”, the better question is “where can a workflow run on its own, with human oversight reserved for the decisions that carry the most risk or the most value?”. Answering that question honestly is what separates organizations that redesign how work moves through the revenue cycle from those that simply layer AI onto the process they already have.
Design for Where You Want to Be, Not Where You Are
The instinct in most organizations is to start with today’s process and look for places to insert AI. The alternative is to start with where the revenue cycle needs to be in three to five years, then work backward. That reordering changes what gets prioritized. Decisions being made today about data architecture, interoperability and workflow design will determine whether an organization can take advantage of the next generation of AI capabilities, or whether it spends the next few years unwinding technical debt instead.
This is also where operational groundwork matters as much as the algorithm. Standardized workflows, clean and accessible data, and systems that expose their core capabilities to software agents, rather than locking them behind a screen a human must click through, are what make autonomous decision-making possible in the first place. An organization that understands its current processes in detail will have a far easier time migrating toward an agentic model than one trying to automate its way out of ambiguity.
Start With the Business Objective, Not the Technology
The most common mistake isn’t technical. It’s deploying AI before defining the business outcome it’s supposed to improve. Every initiative should begin with a measurable objective: fewer touches per claim, higher recovery rates, faster cash flow, or more capacity per analyst. Without a clear target, it’s hard to demonstrate value, build the case for further investment, or know where to expand next.
The stakes for getting this right are rising. Denials have grown sharply across care settings: inpatient denials climbed 220 percent over the past three years to an average of $10,000 per claim, and outpatient denials rose nearly 33 percent to an average of $825 per claim, according to research reported by HFMA. AI needs to be accountable to numbers like these, not the other way around. The organizations that scale AI successfully are the ones that prove the business performs better because of it.
Let the Data Decide Where Autonomy Belongs
The clearest way to sort automatable work from work that still needs a person is to look at the nature of the decision. If the answer already exists in the data, and the decision is consistent, repeatable and governed by known rules, that work is a candidate for autonomy. Claim status checks, portal and voice-channel retrieval, payment posting and routine account resolution fall into this category. People shouldn’t spend their time re-making decisions the organization has already learned how to make.
Human expertise becomes more valuable as uncertainty rises. Negotiation, exception management, ambiguous payer or clinical scenarios, and decisions that require weighing competing priorities are where people create the most value. AI can support that work by pulling claim status from a payer, summarizing a lengthy medical record, or drafting a first pass at an appeal letter, but it shouldn’t be expected to finish it alone.
The current state of adoption reflects that split: HFMA reports that 27 percent of organizations are already deploying AI at scale across multiple revenue cycle functions, while more than half are still piloting in select areas. Most health systems, appropriately, are moving toward autonomy in phases rather than all at once.
The Outcomes That Should Anchor the Work
Whatever the sequencing, the business case has to point somewhere specific. That usually means fewer manual touches per claim, with less rework caused by avoidable errors upstream. It means faster cash realization, from cleaner claims at submission through quicker resolution of denials and underpayments. And it means capacity that grows without headcount growing at the same rate, so staff can spend more of their time on the accounts that genuinely need judgment and negotiation, rather than on the ones that simply need someone to check a box.
None of this argues for fewer people in the revenue cycle. It argues for a different role for them: less time spent repeating decisions the organization already knows how to make, more time spent on the exceptions, negotiations and judgment calls that move the needle. The health systems that pull ahead over the next several years won’t necessarily be the ones with access to the most advanced models. They’ll be the ones that used this moment to redesign how work moves through their revenue cycle in the first place.
