AI Is Making Your RCM More Efficient. That Might Be the Problem.

Your AI vendor just told you they processed 40,000 claims last month. Congratulations — now ask them how many of those claims got paid on the first pass.

That question makes most vendors uncomfortable. And that discomfort is exactly the problem.

Efficiency ≠ Effectiveness

The RCM industry has spent the last two years measuring AI by activity: tasks automated, touches handled, denials worked. What it hasn't done is measure whether any of that activity actually moved money. According to MedEvolve CEO David Henriksen, somewhere between 65% and 85% of human touches in revenue cycle produce zero financial outcome. They're not resolving claims. They're not recovering denials. They're just… work.

Now we're deploying AI to do that same non-actionable work. Faster. At scale.

That's not a win. That's an efficiency trap.

The Touch Tax

Henriksen's framing is blunt and worth stealing: every human or AI touch that doesn't produce payment is a Touch Tax — a cumulative cost you're paying for activity that doesn't advance the claim.

The math is ugly once you run it. Take a $635 claim that goes through six or seven touches at $5–$10 per touch before it gets paid. You've spent $30–$70 in labor to collect $635. Now compare that to the same claim resolved in a single touch. The difference isn't just operational — it's the difference between a practice that's viable and one that's slowly bleeding out.

AI doesn't fix this by default. If your AI system is trained to work every claim in the queue, it will work every claim in the queue — including the ones that will never pay, the ones that required a prior auth you didn't get, the ones that needed a clinical note attached two weeks ago. Faster churning of futile work is still futile work.

What Good Measurement Actually Looks Like

If you're going to hold your AI vendor accountable — or your own RCM operation — these are the metrics that actually matter:

• Touches to resolution — How many touches does it take to close a claim? Not how many were processed.
• Avoidable touches — What percentage of touches could have been prevented with better front-end processes or clean claim submission?
• Denial-related workload — What share of your team's time (and your AI's time) is spent on denials that could have been avoided?
• Payment outcomes per touch — Of every action taken on a claim, how often does the next event result in payment?
• Total cost to collect — Fully-loaded, including AI licensing, staff time, and rework. Not just the software line item.

None of these are exotic. They're the metrics any serious RCM operation should have been tracking for years. The difference is that AI forces the question: if you're automating 10x the volume of touches, you need to be 10x more certain that those touches matter.

Pressure-Test Your AI Vendor

Before you sign a renewal or expand an AI contract, ask these five questions out loud. How they answer — or whether they squirm — tells you everything:

1. "What percentage of claims your system touches get paid on first pass?" — If they only know total claims processed, they're measuring activity, not outcomes.
2. "Can you show us our Touch Tax — the cost of work that didn't produce payment?" — Any serious analytics platform should be able to surface this. If they can't, you're flying blind.
3. "How does your system decide what NOT to work?" — Prioritization away from low-probability claims is as important as working high-probability ones. Does their AI know the difference?
4. "How has our total cost to collect changed since deployment?" — Not just automation rate. Not just denials worked. Total cost. If they don't know, that's the answer.
5. "What's your definition of a successful outcome — and how do you measure it?" — If the answer involves any form of task volume, escalate the conversation.

The Bottom Line

AI in RCM has enormous potential — but only if it's pointed at the right problem. Automating a broken workflow doesn't fix the workflow; it just breaks it faster and at lower cost per repetition. The practices and health systems that will win with AI are the ones demanding outcome accountability from day one: not how many claims your system touched, but how many it resolved, avoided reworking, and got paid cleanly.

Efficiency is easy to sell. Effectiveness is harder to prove. Know which one you're buying.

Source: MedEvolve / PR Newswire