By David Main DC, President, EasyDocForms
LinkedIn: Dr. David Main
Of everything artificial intelligence was going to fix in healthcare, billing looked like the softest target. It is rule-bound, high-volume, repetitive. The code sets are published, the edits are published, the payer policies at least nominally so. If you had asked me three years ago where AI would land first and hardest, I would have said the revenue cycle without hesitating.
There is an alternate universe in which that bet paid off. In that version of mid-2026, the practice’s AI and the payer’s AI read the same chart note, apply the same criteria, and settle the claim between themselves in an afternoon: two machines doing in minutes what now takes two staffs three months. That was a plausible forecast. It is not remotely what happened.
My work puts me inside the back offices of small and mid-sized practices, helping with documentation. A narrow window, but an honest one. What I see is not a settlement. It is an arms race that has reached a tipping point.
I see confusion over code edits. Documentation that does not support what was billed, and staff genuinely unsure what would. Prior authorizations that go out clean and come back denied. And I see payers stalling, delaying and denying claims any reasonable reviewer would pay on grounds that shift depending on who answers the phone. In dentistry it is at its most brazen: routine restorative and periodontal work, thoroughly documented, sitting in limbo for months against a wall of requests for records already sent.
The clearest signal, though, is not what I see in practices. It is what shows up in my inbox.
I get pitched weekly by outsourced billing and RCM firms, dozens upon dozens of them based in India, the Philippines, and elsewhere. If AI had eaten medical billing, that industry would be contracting. It is doing the opposite: Everest Group has tracked RCM outsourcing growing at better than 12% annually, with the outsourced market on pace to roughly double before the decade is out. Hundreds of thousands of people are still being paid, in lower-cost countries, to do the work the machines were supposed to absorb.
I think we mistook the nature of the knowledge. AI is extraordinary with codified knowledge, which is why autonomous coding genuinely works in emergency medicine, radiology, and pathology, where volume is high and inputs are uniform. Those are real wins. But coding was never where most of the labor lived. That lives in eligibility, prior authorization, claim status, denial follow-up, and appeals; and what a veteran biller knows about that work was never written down. It is not the payer’s policy manual. It is what that payer actually does: which carrier reverses on a peer-to-peer, which one needs the narrative attached rather than referenced, which one quietly moved its documentation requirement last quarter. That knowledge is tacit, adversarial, and constantly moving, and it appears in no training corpus.
The other reason is less comfortable. The other side automated first, and better. In the AMA’s 2026 survey, 74% of physicians said denials have risen over the past five years and 60% expect AI to push them higher. Every automated denial manufactures human appeal work downstream. We did not automate the conflict. We scaled it.
And practices are absorbing that while being squeezed from every other direction: labor, rent, malpractice, supplies, software. Reimbursement has not kept pace for years. Adjusted for practice-cost inflation, Medicare physician pay has fallen 33% since 2001. The 2026 fee schedule pairs a 3.26% conversion factor update with a new 2.5% “efficiency adjustment” that cuts thousands of services on the premise that they have become more efficient over time. Read that again. Physicians are being paid less on the assumption of productivity gains that the technology has not delivered to them, in the same year their denial burden went up.
So good clinicians are leaving. Cash-pay, direct primary care, dropping Medicare and Medicaid entirely. On a recent call I learned there were only a handful of private practice dermatologists seeing Medicaid in Ohio; roughly one in ten primary care physicians now practices some form of direct care. That is the real consequence, and it shows up in nobody’s AI adoption metrics.
I am not arguing AI has no place in the revenue cycle. I am arguing we should stop describing a fight as a solution. The only leverage most practices still control is upstream: capture complete information at intake and document the encounter so it supports the claim before the claim is filed. Unglamorous, and the one part of this nobody can outsource for you.
The smart money said billing would be solved by now. On the ground it isn’t, and the people it is failing are starting to walk.