I was on a call recently with a VP of Revenue Cycle at a 300-bed standalone community hospital in the Midwest. Forty-three years in healthcare. Seen the whole arc of it -- from paper charts to EHRs, from fee-for-service to Medicare Advantage. Smart, frustrated, and deeply honest about what's not working.
He walked me through something that stuck with me. His professional coding manager had pulled up a claim that morning with seven different codes on it. Each one needed to be verified against the documentation. Running down the middle of the screen: a column of edits, one for each line. One of the codes was a consult -- but the provider who entered it was actually the patient's attending physician, so the consult wasn't billable. She had to dig into the documentation, change the code, eliminate the edit, and check whether the documentation itself needed updating. Then repeat that for every other line.
That's one claim. His team processes hundreds of them every day.
I asked him where the real problem was. His answer was immediate: "Providers do not want to have to stop and do something -- even though they end up having to do it on the back end anyway."
He's right. And that observation points to a structural flaw that no billing department, no clearinghouse upgrade, and no workflow automation tool downstream can fully fix. The revenue cycle problem in most healthcare organizations is not a billing problem. It's an upstream problem.
Where the money is actually lost
McKinsey estimates that AI-enabled revenue cycle management could cut cost to collect by 30 to 60 percent. Nearly 20 percent of claims are denied on average, and as many as 60 percent of those denials are never appealed -- representing millions in preventable lost revenue sitting on the table every year.
~20%
of claims denied on average
60%
of denials never appealed
$25-118
cost to rework each denied claim
But here's what those numbers obscure: by the time a claim is denied, it is already too late to fix the root cause efficiently. The clinical encounter happened. The documentation was written. The order was entered. The code was assigned. Every one of those decisions -- made by providers whose job is to care for patients, not navigate payer policy -- carries a financial consequence that won't surface until days or weeks later, in a completely different system, reviewed by completely different people.
The physician ordering a procedure doesn't know that specific payer has a 34 percent denial rate for that order with that diagnosis. The hospitalist documenting an admission doesn't know whether the documentation supports inpatient status or gets downgraded to observation. The coder finalizing a claim doesn't have visibility into the fact that this exact claim configuration has been denied by this payer eight times in the last six months. They're operating blind -- not because they're not good at their jobs, but because the information they need lives somewhere they never look.
The training problem no one talks about
The VP I spoke with said something else that stayed with me: "Providers didn't get into healthcare to document."
That's not a complaint. It's a design observation. We have built a system that requires clinicians to simultaneously be caregivers and billing compliance officers -- and we train them for one and expect them to perform at both. When DRGs arrived in the mid-eighties and payment became tied to coding, the entire incentive structure of documentation changed overnight. But the way we train, support, and hold providers accountable barely changed at all.
The result is a workforce that has adapted by pushing through alerts, copy-pasting notes, and using workarounds -- not because they're careless, but because the system isn't giving them the right information at the right moment in a way they can act on. When a provider sees an edit that says "stop -- you don't have a valid diagnosis for this order," they blow past it. Not because they don't care. Because they have a patient in front of them and the edit doesn't help them understand what to do differently.
The path to cleaner claims runs through clinical workflows, not billing workflows. The data problem has to be solved at the source -- before the claim exists.
What moving upstream actually means
I've spent the last several years building AI-powered tools inside healthcare organizations -- primarily in post-acute care -- and the pattern I keep seeing is the same. The organizations that make the biggest dent in their revenue cycle performance aren't the ones with the best denial management software. They're the ones that have closed the gap between the clinical decision and the financial consequence.
That means surfacing the right information at the point of care -- not as a hard stop, not as a compliance burden, but as context. A soft layer of intelligence that makes the better path the easier path. It means connecting what a provider documents today to what gets paid six weeks from now, and making that connection visible before the claim is filed.
We built exactly this kind of upstream intelligence layer for one of the nation's largest post-acute care providers. The work centered on MDS assessments -- a clinical process that directly drives PDPM reimbursement. By embedding AI into the clinical workflow, we helped their teams identify assessment discrepancies before they translated into billing errors. The result was over $10 million in recovered PDPM revenue and a 95 percent reduction in assessment discrepancies -- not by adding more billing staff, but by fixing the data problem at the source.
Similar work in RAF/HCC coding for a Medicare Advantage population produced $12 million in recurring annual revenue by surfacing coding anomalies before the risk adjustment window closed. In both cases, the ROI was measurable, near-term, and validated by independent financial analysis. That matters enormously to CFOs and COOs who are being asked to invest in AI at a moment when margins are thin and skepticism about technology promises is high. The question they need answered isn't "is this interesting?" It's "how fast do we get paid back, and how do we know?"
This isn't a technology problem. It's a sequencing problem.
The tools to fix the upstream revenue cycle exist. The AI frameworks for denial prediction, coding accuracy, payer contract interpretation, and clinical documentation improvement are mature enough to deliver meaningful ROI in eight to twenty weeks. What's missing in most organizations isn't access to technology -- it's a clear-eyed decision about where to start and how to sequence the investment to generate early wins that fund the next phase.
Throughout this series, I'll dig into each dimension of that challenge: what payer-aware intelligence actually means and why org-specific models outperform generic tools; how to make clinical workflows payer-aware without destroying provider adoption; why payers have industrialized denial and what it takes to fight back systematically; and how to build an RCM AI roadmap that the CFO will actually approve.
But the starting point is recognizing that the billing department can't fix a problem that originates in the exam room. The data has to get clean before the claim is built. That's the upstream imperative -- and it's where the real money is.
