AI was sold to health care as a way to reduce paperwork and eventually lower costs. One of its first visible effects is helping providers bill more for the same encounter.
The uncomfortable part of the latest health care AI story is not that the technology failed. It may be working exactly as designed. PwC’s new 2027 medical cost outlook, published June 11, says commercial medical cost trend is expected to hit 9% next year, the highest level in 17 years. In the individual market, the projection is 8.5%.
Those numbers are not only about expensive drugs or hospital wages. PwC names provider adoption of AI-enabled revenue optimization tools as one of the five main inflators, alongside reimbursement pressure, pharmacy costs, behavioral health utilization, and No Surprises Act arbitration. The firm said 70% of health plans rank AI-enabled tools that capture more provider revenue as a top-three cost inflator.
That is a very different story from the one most AI vendors prefer to tell. The pitch has usually been about freeing doctors from screens, shortening the time spent writing notes, and removing administrative waste. All of that can be real. But in a system where a more detailed note can support a higher code, better documentation can quickly become better monetization.
PwC’s report puts the mechanism plainly: providers are using AI-enabled documentation and coding tools to record greater specificity and reimbursable severity, which means payers see higher paid amounts per claim. Axios, which reported on the PwC outlook on June 12, framed the problem as a structural one. Health care is still largely paid to do more and bill more, so AI does not enter a neutral system. It enters a system with a very clear business model.
That is the point founders and investors should sit with. A startup that helps a hospital code more precisely may describe itself as a productivity company, but the buyer may experience the product as a revenue tool. In health care, the difference is not academic. It decides who saves money and who receives the next larger bill.
Blue Cross Blue Shield made the same issue harder to dismiss in March. Its Blue Health Intelligence analysis looked at de-identified claims from tens of thousands of maternity admissions and found sharp increases in acute posthemorrhagic anemia coding at some hospitals, without a matching increase in transfusions. The analysis said the rise in that one diagnosis added $22 million to maternity admission costs in one year.
The Blue Cross report went further. It said about $663 million in inpatient spending may be tied to AI-powered coding tools, and at least $1.67 billion in outpatient spending may be linked to more aggressive coding practices. It also cited a 2023 HFMA and AKASA survey finding that 46% of hospitals and health systems used AI in billing, coding, and claims, while federal data showed AI use for billing jumped 25 percentage points year over year.
This is not proof that every AI-coded claim is wrong. Better documentation can reveal real complexity that was previously missed. A patient with multiple conditions should not be flattened into an incomplete record because the paperwork was annoying. But when coded severity rises faster than the underlying care being delivered, insurers and employers are right to ask whether software is improving accuracy or just improving collections.
The cost fight moves to employers
PwC’s survey covered actuaries at 27 US health plans, representing more than 103 million employer-sponsored members and 8 million individual ACA marketplace members. That makes this a boardroom issue, not only a hospital finance issue. Self-funded employers will feel these changes through claims experience, vendor oversight, benefit design, and the yearly fight over what workers can still afford.
There is already a familiar counter-move forming. PwC says payers should improve coding-intensity surveillance, severity-shift monitoring, payment integrity, and pre-payment review of high-dollar claims. The goal, as PwC phrases it, is more accurate payments, not more denials. That distinction matters, because insurers have their own AI problem. UnitedHealth Group and Cigna have both faced lawsuits over alleged algorithmic claim denials, a reminder that automation can be used on both sides of the bill.
Axios also cited a UBS report that looked at AI through the financial position of insurers and hospitals. The insurer side is likely to become more efficient, but those gains may be competed away if every carrier buys similar tools. Hospitals are not equal either. Large for-profit operators such as HCA, Tenet, and UHS have more capital and operational capacity to adopt AI quickly than many nonprofit systems.
The more hopeful version of this story still exists. Harvard Medical School associate professor Hossein Estiri told Axios that attention may move away from administrative AI and toward tools that improve patient outcomes. Venrock partner Bob Kocher made the sharper economic point: if AI is applied inside fee-for-service medicine, it will likely increase units of care and coding. If payment models reward lower total cost, the same technology could be pushed in a different direction.
For now, the evidence points to a more immediate conclusion. Health care AI is not naturally deflationary. It follows the incentives in front of it. In 2026, many of those incentives still point toward more documentation, more severity, and higher paid claims before they point toward cheaper care.
Also read: Nectar Social shows where enterprise AI is moving next • New research finds AI data center water fears are outrunning the actual science • Apollo's hunt for a second headquarters signals a permanent shift in where private equity capital calls home
