AI in the Revenue Cycle: It’s Already There. Govern It.

Artificial intelligence is already operating inside your revenue cycle. 

It is in your clearinghouse’s claim edits. It is in your payers’ prior authorization engines. It is in the coding prompts built into your electronic health record. So the strategic question is no longer whether to adopt AI in revenue cycle management. It is whether you can see where it touches your claims, and whether you can prove the right controls sit around it. The practices that win will not be the ones that moved fastest. They will be the ones that paired speed and scale with governance, validation and human review.

An honest place to start

Let me start with a confession. I once asked an AI tool to write a birthday note for my mother. It produced a line about her golden years continuing to glow. My mother spent her career calculating kill ratios for missile systems. She was not charmed.

That is the whole story of AI in two sentences. Incredibly capable. Also fully capable of misjudging the room. The gap between what these tools can do and what they do without oversight is exactly what every healthcare leader needs to understand before signing anything.

Why this matters for your revenue

For gastroenterology and rheumatology practices, the stakes are not abstract. Your revenue rides on biologics and infusions that almost always require prior authorization, and prior authorization is where dollars stall.

93% of physicians say prior auth delays care (AMA, 2024)

39 prior auths per physician, every week

72 hrs new CMS urgent-decision limit by 2026

In the 2024 AMA Prior Authorization Physician Survey, 93 percent of physicians said prior authorization delays care, and practices reported completing an average of 39 authorizations per physician each week. Every delayed authorization is a delayed or canceled appointment. Every preventable denial is cash that ages in accounts receivable or never arrives.

The rules are changing, too. Under the CMS Interoperability and Prior Authorization final rule, impacted payers must return prior authorization decisions within 72 hours for urgent requests and seven calendar days for standard ones, with electronic processes phasing in through 2027. Faster payer decisions only help you if your submissions are complete and your denials are worked with discipline. That is precisely where well-governed AI earns its keep. It surfaces missing requirements before a request goes out. It flags high-risk claims before they are submitted. It classifies the root cause when a denial does land.

What AI actually is, and what it isn’t

Strip away the movie version. AI is not self-aware, it is not plotting anything, and it is not coming for your staff. It is a very sophisticated pattern matcher. It predicts. It does not think.

The most useful way to picture it is a new employee on day one. You hand them every policy, claim form, payer contract and denial letter your practice has ever produced. They read all of it. They never tire, never call out on a Monday, and can work thousands of cases at once. There is one catch. Now and then they will tell you something that is simply wrong, not out of dishonesty but because they are predicting from patterns. In healthcare, we call that a hallucination, and it is the entire reason governance is not optional.

Where AI already touches your claims

AI is not a future purchase. It is already running inside the systems you use today.

• Your clearinghouse. Claim scrubbing and edit logic at major clearinghouses is increasingly AI assisted. Your claims already pass through it.
• Your payers. Prior authorization engines use AI to approve or flag requests. Understanding that logic, and its gaps, changes how you fight a denial on a biologic or an infusion.
• Your EHR. Epic, Oracle Health and athenahealth are all building AI assisted coding suggestions and documentation prompts directly into the workflow.
• Your RCM partner. Any partner worth evaluating should be able to show you exactly where AI touches their work and what governance surrounds it. If they cannot, that is your answer.

What changes, and what doesn’t

AI does not change the goal. It changes the speed, the scale and where the risk lives.

Today, without governance	With governed AI
Staff submit a prior auth and wait. The denial returns 14 days later. Appointments get moved or canceled.	Missing requirements are flagged before submission. The package goes in complete the first time.
A denial hits the remittance. Staff work a queue. Some appeals get filed, some do not.	High-risk patterns are caught pre-submission. Root cause is classified and the appeal is structured automatically.
Compliance rules live in someone’s head, a 300-page manual, or sticky notes around a monitor.	NCCI edits, MUE limits and LCD coverage are enforced in code, on every claim, with an audit trail.

What good looks like

Here is the standard to hold every vendor to, including your own teams. Ask for evidence, not promises.

• Human review is non-negotiable. No AI should route a coding or clinical decision without a person in the loop. Ask whether that is policy or whether it is enforced in code. There is a difference.
• Compliance is verifiable. NCCI edits, MUE limits and LCD coverage should be enforced on every claim, and a vendor should be able to show you the tests, not just the slide.
• Audit trails are complete. If AI touches a claim, you should be able to trace every recommendation and every decision. That is what defensibility looks like in a payer audit.
• Failure modes are safe. When the tool is wrong or offline, the answer can never be that something silently passed through. Mistakes will happen, with software and with people. Safe-mode fallback is a requirement, not a feature.

Speed  +  Scale  +  Guardrails  =  Trust Speed and scale without guardrails is just liability moving faster.

Three camps. Where are you?

Most leaders fall into one of three camps, and two of them are expensive.

The first camp believes AI fixes everything. Deploy it, watch denials vanish. The risk is no guardrails, real compliance exposure and false confidence until an audit arrives.

The second camp believes AI is too risky and decides to wait. The risk there is quieter but just as real. Your competitors are not waiting, and your denial rate certainly is not. Many of them have already tried, failed and rebuilt.

The third camp treats AI as what it is: a powerful tool that has to be built right. Speed and scale, governed by validation, audit trails and human review. The architecture around the tool is what makes the speed trustworthy. That is the camp worth being in.

A checklist for practice administrators

Before you sign with any vendor, new or existing, put these five questions on the table.

1. Show me the map. Where exactly does AI touch our claims, and where does a human stay in the loop?
2. Show me the tests. How are NCCI edits, MUE limits and LCD coverage enforced and verified, not just described?
3. Show me the trail. Can every AI recommendation and decision be traced for a payer audit?
4. Show me the fallback. What happens when the AI is wrong or offline?
5. Show me the proof. Not the marketing deck. The controls.

If a vendor cannot answer those, you have learned something valuable for free.

We’re just getting started. AI in the revenue cycle is early innings, and the rules are being written now. That is exactly why the standard is defensibility: audit trail, human review and evidence you can verify independently. At Advantum Health, that standard is built into how we manage prior authorization and eligibility verificationand denial management for specialty practices. See where AI already touches your revenue cycle, and what governance should sit around it. Let’s talk.