Blue Cross Estimates AI Hospital Coding Added $942 Million in Costs

Hospitals' use of AI-assisted medical coding added an estimated $942 million in costs for Blue Cross Blue Shield Association health plans between 2023 and 2025, according to the insurer's own analysis. Roughly 70% of that amount — $653 million — was tied to additional diagnoses that came with no accompanying change in patient care.

The finding lands at the center of a fast-moving shift in U.S. healthcare: AI now shapes not only diagnosis and treatment but what hospitals bill, what insurers pay, and which claims get denied. And the early evidence challenges the assumption that AI in administration will act as a deflationary force in an already expensive system.

What Blue Cross Blue Shield found

BCBSA said much of the cost increase came from secondary diagnoses that moved patients into higher-paying reimbursement categories — diagnoses that "may be derived from single laboratory values, making it particularly well suited for detection by AI tools."

The growth in what BCBSA calls "complex coding" occurred during a period when 60% of hospital systems began using AI coding tools, the association wrote in its analysis. "There is a clear disconnect between coding and treatment," the report stated.

Luke Chalker, BCBSA's senior vice president of product and data science, stopped short of attributing the entire increase to AI. "While multiple factors contribute to coding intensity, the findings suggest AI-enabled coding and documentation tools are playing a role," he told CNBC.

He said consumers have reason to be concerned. More complex coding can drive higher reimbursement without more care, and "those costs can eventually show up in the form of higher premiums and out-of-pocket costs."

The stakes are concrete. Benefits consulting firm Marsh forecasts the cost per employee for health coverage will rise 8.2% on average in 2027 — which would mark the highest increase since 2003.

The hospitals push back

The American Hospital Association rejected the insurer's framing. "Patients today are older and more clinically complex," an AHA spokesperson said in a statement, and AI tools are helping providers "appropriately capture their patients' conditions to aid in care planning. The BCBSA's analysis lacks the context needed to meaningfully assess how these tools impact healthcare quality, patient access, or spending."

The spokesperson turned the accusation around: "It is particularly troubling to see insurers raising concerns about provider coding while continuing to rely on automated downcoding and denial practices that can impede coverage of medically necessary care, add burden on the workforce, and increase costs through administrative waste."

Chalker acknowledged that Blue Cross Blue Shield companies also use AI in claims review, but said "any clinical denial is always reviewed by a qualified human clinician."

An 'administrative arms race'

Christopher Whaley, a health economist at Brown University who studies hospital coding, said AI appears to be "accelerating, and in some sense making it easier to capture, the existing and underlying billing incentives that are in the system."

The diagnoses themselves are not necessarily wrong. "In many cases, the diagnoses are legitimate and weren't captured," Whaley said. But he noted there are also conditions that "clinically just don't really matter and don't influence the patient's care," while still allowing another billing code and increasing payment.

With sophisticated AI deployed on both the hospital and insurer sides, Whaley warned of an "administrative arms race." Those tools "are both very expensive," he said, "and also have nothing to do with providing appropriate care to patients." The costs ultimately flow to consumers through higher premiums and taxes.

Marisa Greenwald, a partner at Oliver Wyman's Health and Life Sciences practice, which advises hospitals and insurers on AI adoption in revenue-cycle management, said the technology is already delivering real benefits. It cuts after-hours documentation work and gives doctors "some semblance of work-life balance back." There is a financial upside too: physicians with more time can see more patients, and more accurate coding can "catch additional acuity and diagnosis components," producing higher reimbursement.

Still, Greenwald said it is hard to separate genuine documentation improvement from overcoding. "It's hard to disentangle how much of it is better accuracy. There's always going to be misuse and user error and overcoding," she said. If insurers respond to provider AI with denial-driving AI of their own, "the arms race is poised to exacerbate. The hope is going to be that on both sides of the equation, the players recognize that all we're doing is adding cost and burden into an already challenged and belabored system."

Otherwise, she said, healthcare could end up with "robots talking to robots and just fighting with each other."

Old tensions, new tools

Vanessa Moldovan, author of "The Healthcare Revenue Cycle AI Playbook" and head of RCM strategy at Magical, which builds AI-powered automation software for healthcare administration, said the underlying tensions predate AI. "The AI is new, but the rest of it is not new," she said. "You could line up 10 coders and have them all look at the same chart and they could code it differently."

AI lets provider organizations review far more records far more quickly, which can help them get paid for care they actually delivered, Moldovan said. But she drew a firm line at diagnoses unsupported by the medical record: "If the patient didn't have those conditions, they didn't have those conditions." For that reason she opposes fully autonomous coding. "There should always be a human in the loop."

She said AI-generated coding should be audited the way healthcare organizations have always audited human coders. Chalker agreed, saying AI should "support decision-making, not replace human judgment."

The risks extend beyond payment. Diagnoses become part of a patient's permanent medical record, and Moldovan worries about misplaced confidence in machine output. "I think we're in danger of trusting it too much because we're like, 'Oh cool, it's AI, it must be smarter,'" she said.

AI now sits on multiple sides of the same dispute: it helps providers code care, helps insurers scrutinize claims, and helps providers challenge denials. "We can get more claims out. We can get more appeals out," Moldovan said.

The real test

Whaley argued the ultimate measure should not be whether AI lowers spending. If the technology increases costs while improving access or quality of care, that could still be worthwhile, he said.

"If AI tools are just used to kind of shuffle the cards a little bit more and make sure you come out on top and not do anything to patients," Whaley said, "then that's something that probably isn't worth investing resources in."