E-News Exclusive

Health Care’s Quest for a Clean Claim

By Selena Chavis

Experts weigh in on the power of AI and predictive analytics to improve denials.

Revenue cycle professionals have prioritized claims accuracy and compliance as a process improvement priority for decades. Many agree that while the challenges associated with submitting clean claims haven’t changed much, they have certainly gotten more expensive.

According to Raja Yogeshwarar, executive vice president of growth and strategy at ECLAT Health Solutions, denials resulting from documentation that doesn't fully support the billed code level, medical necessity gaps, missing prior authorizations, and coordination-of-benefits errors remain the usual suspects. The less obvious problem is inconsistent coding patterns across providers or departments that don't look like bad claims individually but eventually surface as a trend a payer algorithm flags.

"The problem is that, traditionally, health systems find out about them in the denial or the audit letter, months after the patient encounter, when the medical record is cold and the staff who documented it may not remember the details," Yogeshwarar says.

One area of claim denials that has changed is the speed and scale of payer response, says Dana Finnegan, senior director of market strategy at MDaudit. "Payers deploying their own AI [are] really doubling down on denials," he says, "and the other thing is a massive increase in external payer audits as well."

Notably, MDaudit's 2025 benchmark data backs the assertion that denials and provider costs are rising. The average denied amount across hospital inpatient and outpatient settings rose in double digits—14% outpatient, 12% inpatient—while the total at-risk amount and case volume for external payer audits climbed 30%, with the average dollar amount per claim up 18%. The findings are compiled from the first three quarters of 2025 across more than 4,500 facilities, 1.2 million providers, and $20 billion in annual charges audited, spanning 20 payer types in 45 states.

The good news is that health systems and other provider organizations can leverage AI and predictive analytics in similar ways to get in front of issues. Rather than waiting for a denial or audit letter to reveal a problem months after the fact, providers now have the opportunity to flip the sequence. By using the same claims and remittance data payers are mining against them, predictive analytics can help catch documentation gaps and coding risk before a claim is submitted.

AI Spots Risk Before Submission
Finnegan says, historically, denials clustered around authorization and eligibility issues, but today the target is "coding-related denials—that's really what they're going after at this point." MDaudit's data shows a 26% year-over-year increase in coding-related denials across professional and hospital outpatient settings in 2025, continuing to build on a 126% increase the organization reported in 2024.

Describing the underlying mechanics of predictive models, Yogeshwarar notes that “it’s pattern matching at a scale humans can’t do manually.” The models are trained on historical claims, drilling down into those that were denied, downcoded, or flagged in audits to learn what those problem claims have in common. Then, on a new claim, AI and predictive analytics compare factors such as: Does the documentation language match what's typically present when this code combination gets paid cleanly? Is this procedure/diagnosis pairing statistically unusual? Is there a mismatch between the acuity implied by the code and the acuity described in the note?

“None of this is the model understanding medical necessity the way a physician does,” Yogeshwarar explains. “It's recognizing that this claim looks statistically similar to claims that caused problems before, and it's flagging it for a human to actually review.”

Finnegan offers a concrete example: An engine trained on a payer's own historical denial data might catch that a claim needs its principal and secondary diagnosis codes resequenced before it goes to a specific payer. He calls it a closed-loop process using retrospective denial data to inform what gets flagged prospectively.

Different than the static, rules-based scrubbers HIM departments have relied on for years, Yogeshwarar says that “a predictive model tells you a technically compliant claim is still statistically likely to get denied or scrutinized based on real historical outcomes."

Finnegan is direct about the risk of AI nudging behavior in the wrong direction, pointing specifically to a recurring failure pattern where "set it and forget it" autonomous coding models often drift into coding intensity—quietly upcoding without anyone catching it in real time. "Nothing is ever autonomous,” he emphasizes. “There always has to be a human in the loop to QA [quality assurance] it, govern it, and help that engine get smarter."

What's Next
Finnegan and Yogeshwarar believe that AI’s impact on claims will ultimately evolve to real-time feedback at the point of care, where documentation is checked against predictive rules before a claim is even generated. Yogeshwarar describes a physician or clinical documentation specialist getting a prompt in the moment that might say, "This diagnosis is documented, but the supporting detail a payer will look for isn't here yet." In this scenario, the fix takes 10 seconds, instead of a query three weeks later.

While Finnegan sees this as the holy grail, he cautions that the reality of this advancement is still years away, and he warns against outrunning the guardrails to get there. "We have to be careful not to move too fast, because that's when it's going to result in coding intensity, drift, up-coding, penalties, and claw backs."

— Selena Chavis is senior director of accounts with Insenna and a Florida-based freelance writer.