What Your Clean Claim Rate Might be Hiding

Hi and happy Tuesday,

Earlier this year, we had a back-and-forth with the operations director at a large multi-site urgent care network managing thousands of patient visits daily.

When we shared a case study about a group that saved nearly $700K by moving from manual claim scrubbing to AI-assisted workflows, his reply was blunt:

“Why did they have a manual claim scrubbing process to begin with?”

Fair point. This network didn’t do things manually. They built their own tools.

They had it all (or so it seemed)

This urgent care network had invested heavily in building their own internal infrastructure:

• Engineers exposing APIs across multiple vendors,

• Custom-built tools linking EHR, charting, and billing

• A centralized team reviewing claims before submission

From the outside, this was a sophisticated operation. From the inside, nonetheless, small issues were slipping through, and piling up, and up, and up.

What did their in-house AI tool miss?

Even with strong systems, they struggled with:

• Unsigned charts — notes written but not finalized
  

• Missing codes — mentioned in documentation, absent from billing
  

• Inconsistent inputs — assistant-added codes clashing with provider notes
  

• Narrative gaps — reviewers missing context unless they read every word
  

Here’s the irony: most of the “automated” processes were still relying on humans to interpret context and correct documentation gaps before submission. Which begs the question...

What does “automation” actually mean?

Not all automation solves the same problems. Here’s how the models compare:

The take-home message: Even the best internal systems still rely on humans to interpret medical notes line by line. That’s the bottleneck.

The blind spot few RCM teams talk about

Everyone focuses on speeding up claim submissions. But the real performance gap often shows up after a denial:

• Denials take days (or weeks) to reach the right person
  

• Staff waste hours chasing documentation
  

• Appeals get buried in inboxes or spreadsheets
  

This is where newer forms of AI show real promise, not just checking boxes, but learning from denial patterns and making the right suggestions upfront.

Why this story matters

The urgent care group we spoke to didn’t want help. They didn’t need it. They’re the exception.

But most practices don’t have engineers on staff or the time to maintain internal QA tools. If you’re in that camp, there are still ways to get the benefits of smarter review, without the internal buildout.

How to map your RCM blind spots in 15 minutes

Here’s a simple framework we’ve started using with practices to identify where things look automated, but really aren’t.

Ask your team (or yourself) these 5 questions:

1. Which step in your claims process still requires someone to “double check” before submission?

   (Hint: This is often where automation ends.)
   

2. When a denial comes in, how quickly can you trace it back to the originating chart?

   (Long delays = poor feedback loops.)
   

3. How often do different staff members enter conflicting codes for the same visit?
   (This signals inconsistent documentation standards.)
   

4. Do you have a list of payers whose denials have been increasing lately?
   (Payer drift happens silently.)
   

5. What percent of your clean claims go out without someone eyeballing them first?
   (That’s your real automation rate, not what the vendor promised.)
   

This mini-diagnostic doesn’t require any tools, just 15 minutes and honest answers. This is a powerful way to spot the gap between “we’ve automated this” and “this still needs humans.”

Next week, I’ll be sharing some tips and tricks on how to lower your claim denial rate.

See you next Tuesday,
Dino Gane-Palmer