Our Take
The real constraint isn't AI capability—it's that medical coding requires physician intent, not just clinical facts, and that boundary won't move with better models.
Why it matters
Healthcare revenue cycle startups pitch full automation as inevitable. Mayo Clinic's Todd Manion, leading operations at one of the largest integrated health systems, is publicly drawing a line. That matters because Mayo has both the scale and the leverage to push back against vendor claims.
Do this week
Revenue cycle leaders: map your workflows by constraint (clinical ambiguity vs. repetitive admin), then tell your AI vendor which bucket your problem lives in before signing.
Mayo Clinic deploys AI, but only in narrow lanes
Todd Manion, revenue cycle chair at Mayo Clinic, told attendees at the HFMA annual conference this month that while AI is proving useful in his operation, full automation of healthcare revenue cycle work is unrealistic. The reason: clinical documentation doesn't compress into the structured data that automated systems require.
Manion described a concrete example. A physician documents a patient's condition as "pulmonary infiltrate" rather than "pneumonia," even though the patient received all treatments and medications for pneumonia. A coder cannot bill based on clinical evidence alone. Only an explicit diagnosis from a clinician, entered in a specific part of the medical record, appears on a claim. "Unless that diagnosis is in a specific place, we can't apply it to the claim without then going back to the provider and querying," Manion said.
At Mayo, AI is already in use—but only in repetitive, low-judgment work. The system checks claim statuses with payers, flags outstanding remits, and follows up on payments exceeding contractual timelines. Tasks that once required staff to hold on the phone with payers now run automatically. "I don't need people waiting on hold to figure out where a claim's status is with the payer," Manion noted. Those freed staff members move to work requiring actual judgment.
The startup pitch doesn't account for billing's legal structure
Much of the AI-in-healthcare-revenue-cycle narrative assumes the problem is execution speed: claims follow rules, payers follow guidelines, most work is repetitive. That framing is incomplete. It misses that medical billing is legally constrained by what a physician explicitly documented, not what the clinical record contains.
Manion's framing resets the conversation. His goal is "to accurately reflect the care that was actually delivered." That's not a technology problem. It's a governance and documentation problem. Better algorithms won't close the gap between what physicians observe and what they document in the right format at the right time.
This distinction matters because Mayo Clinic has both the data maturity and the operational sophistication to use AI where it fits. If Manion is skeptical of full automation, it's not because Mayo lacks AI capability or courage. It's because he understands the actual constraint.
Separate automation from judgment-lifting
If you're evaluating AI for revenue cycle, stop asking "What can we fully automate?" Start asking: "Where is our bottleneck a human waiting instead of deciding?" Mayo's playbook targets phone holds, status tracking, and payment lag follow-ups. Those are high-volume, low-variance workflows. Claim denials tied to coding ambiguity are not.
Manion's second point matters equally: automation should elevate your people toward complex work, not eliminate headcount. That framing tends to meet less resistance from teams and unions, but more importantly, it's honest about where human judgment still creates value.