There’s no question that the price of healthcare continues to trend upward in the United States, with report after report pointing to sustained, elevated costs of receiving care. What’s less clear, though, is exactly what is driving up costs.
Various explanations have been floated over the years, with insurers and health systems repeatedly pointing the finger at one another and, sometimes, at patients themselves. The latest salvo in the ongoing battle between health insurers and hospitals came in late September, with the release of a white paper by Blue Cross Blue Shield insurers examining the use of artificial intelligence in hospital billing.
The paper, which was released Sept. 24, claimed that hospital systems’ billing practices have driven $942 million in added healthcare costs over a two-year period for the Blue Cross Blue Shield system, which comprises 33 independent companies. Blue Cross insurers said the increase stems from secondary diagnoses added to a claim that moves it into a “higher-paying category.”
“The rise coincides with hospitals adopting new AI-assisted billing software that scans records and lab data for anything that can be coded,” the Blue Cross Blue Shield report stated.
Per the report, the share of cases that were billed as “complex” ticked up from 37% in 2023 to 40% by late 2025.
“In one closely studied area, major bowel procedures, the most complex claims went up while the simplest ones fell, even though the associated treatments themselves did not change,” the report said. “The added complexity is concentrated in a subset of hospitals within peer groups (e.g., within teaching hospitals), which points to differences in coding practices more than a broad change in how sick patients are.”
An analysis of diagnoses, the report said, “reveals the fingerprints of the AI-enabled revenue cycle management technologies.”
For me, not for thee?
It’s worth putting some context around the Blue Cross paper. Health insurers themselves, for years, have been deploying AI and other algorithms in their own operations. UnitedHealth Group, the nation’s largest insurer, earlier this year said it plans to spend as much as $1.5 billion on AI-related initiatives. The company also faces a class action lawsuit alleging it relied on AI tools to deny medical claims for Medicare Advantage plans; the insurer disputes the allegations.
On a page for Blue Cross Blue Shield of Illinois, meanwhile, the company said it uses AI for customer service, claims processing and “proactive health support.”
“AI can predict when a member might be at risk for serious health issues, like diabetes or hospital readmission. This helps us reach out early and offer extra support,” the page said.
In an email to CFO.com, Blue Cross Blue Shield Association Senior VP of External Affairs David Merritt acknowledged that “AI has the potential to improve healthcare administration and efficiency.”
Still, he said, “hospitals and payers should ensure diagnoses and billing accurately reflect a patient’s condition and the care that’s being delivered.”
“Employers across the country are struggling with the rising costs of offering coverage, with CFOs often tasked with navigating this incredibly challenging time,” Merritt said.
And despite the latest hype surrounding the latest AI tools, computer algorithms are not really anything new in healthcare. Stephen Parente, a finance professor at the University of Minnesota’s Carlson School of Business who studies health economics, noted that the advent of diagnosis-related groups, or DRGs, in the 1980s marked the first time that computer algorithms made a significant impact on the way hospitals get paid for services.
Speaking about the use of AI in billing today, Parente said that “on the one hand, it’s new, because it’s AI. On the other hand, it’s not.”
Health system leaders say it’s only fair for their organizations to use the technology in turn. McLaren Health Care CFO Dave Mazurkiewicz last month told The New York Times that “all insurers are using AI to scan our charts to look for claims to deny.”
“For the same reason,” he told the newspaper, “we’re looking at the same charts today.”
Room for common ground
One could, of course, argue that not all AI use in healthcare is necessarily a bad thing. If a patient can get access to a provider or life-saving drug sooner by algorithm, few would complain. And, with the prospect of a shortage of tens of thousands of physicians over the next decade, automated tools could, in theory, help address very basic care needs or triage to human providers.
Lalithya Yerramilli, senior VP of payment solutions at Cohere Health, said her firm uses AI to sort through claims, but that the technology will never deny a claim without human intervention. The company currently sells its automatic prior authorization technology only to insurers, for now.
She said as many as 85% of claims run through her system are automatically approved. Those that aren’t are forwarded on for human review.
“Our clinical intelligence platform takes that clinical document, scrubs the intelligence from there, converts it into decision-making, and then auto-approves the auth,” she said.
When the company’s technology finds that “something is missing or something that’s not aligned to the medical necessity policy, we don’t auto-deny it,” Yarramilli said. “We don’t use AI for that.”
As for the cost claims in Blue Cross Blue Shield’s report? “My two cents on it: I don't think it's $900 million, but it's not zero either,” she said. “It's somewhere in between.
She acknowledges, though, that there has been an increase in “upcoding,” which she said happens when providers assign a higher intensity code to describe otherwise normal conditions. “We, as an entire population, just in a short two-year time frame, could not have become sicker,” Yarramilli said. “What we are trying to get to is payment accuracy.”
The concept of payment accuracy would seemingly be appealing to insurers and providers. With both sides racing to implement AI, could they find common ground on a single technological solution? Perhaps. Parente with the University of Minnesota points to the widespread use of FICO credit scores among financial institutions as an example of industry-wide cooperation.
“We basically take the FICO score as the authority, as opposed to the bank making an arbitrary decision,” he said. Banks realized it was better to cooperate in this area than to lose cash to fraud, he added.
But, Parente noted, “the insurers aren’t there” yet.
“Insurers make money reselling data for pharma companies, for drug surveillance and all sorts of other stuff,” he said.
And as both sides continue to duke it out, now in public venues and white papers, patients may ultimately stand to lose the most.
“The biggest concern for me is the patients caught in between,” Parente said.
