The Enemy Is Us. And AI Won't Be the Hero.
We keep looking for someone to blame for the rising cost of healthcare, and we keep hoping technology will ride in and rescue us. I understand both instincts. They are human, and they are comforting. But after most of a career spent inside the revenue cycle, I have come to believe they are two of the biggest things standing between us and the work that actually moves the number.
Let me explain what I mean, and why I think 2027 is the year this stops being an abstract argument.
The search for a “bad guy”
Every few months, a new survey comes out asking Americans who is responsible for what they pay for healthcare. The latest one, reported in June, found that 47% blame health insurers, 36% blame the federal government, and 34% blame drug companies.[1] It is worth knowing that the survey was commissioned by a coalition backed by hospitals, which probably explains why insurers landed on top. Right on schedule, AHIP, the insurance industry's main lobbying group, fired back that hospital care accounts for more than 40 cents of every premium dollar, and suggested hospitals look in the mirror before pointing fingers.[1]
I have watched this fight for most of my career, and I am going to say something that does not make for a good headline. There is no villain here.
A villain is a satisfying thing to have. A villain is simple, and a villain can be defeated. If insurers are the problem, you regulate insurers. If hospitals are the problem, you regulate hospitals. The trouble is that rising cost is not the work of a bad actor. It is the sum of a lot of rational actors, each one doing exactly what its own incentives reward.
Consider the pieces. Hospitals are buying up physician practices at a remarkable pace; by 2024, 47% of physicians worked for a hospital or health system, up from under 30% in 2012.[2][3] They do this partly to survive, because scale helps them weather thin margins. But when a hospital absorbs a practice, the prices for the same office visit tend to rise; one analysis ties that consolidation to a 17% increase.[2][3] Insurers, for their part, deny and delay because that is how they manage financial risk, and the system rewards them for it. Vendors sell documentation and coding tools because the tools work and clients want them. Every one of those decisions is defensible on its own. Add them all together, and you get the number PwC just published: a projected 9% jump in commercial medical costs for 2027, the steepest in 17 years, on a path toward $9 trillion in national health spending by 2035.[2][4]
That is the uncomfortable truth a villain story hides. The cost problem is not a fight between good guys and bad guys. It is a system where everyone is behaving sensibly and the combined result is a number nobody wants and nobody will claim.
The hope that AI will be the hero
If there is no villain, surely there is a hero coming. For the last couple of years, the story we told ourselves in healthcare finance was that artificial intelligence would be the thing that finally bent the cost curve. Automate the busywork, clean up the claims, and watch the savings roll in.
Then PwC released its 2027 outlook and ranked AI-enabled documentation and coding as the single largest cost inflator for the year. Roughly 70% of the health plans it surveyed put it in their top three.[2] Read that again, because it matters. The technology we were counting on to lower costs is now the number-one thing driving them up.
Here is the mechanism, in plain terms. Ambient AI scribes (tools that listen to a clinical visit and draft the note) and AI coding assistants document care in far more detail than a rushed human would. More detail supports higher-severity codes, and higher-severity codes mean higher payment, even when the underlying care did not change. A study of ambient AI scribes at UCSF Health, published in JAMA in January 2026, found higher relative value units per encounter and per week after adoption, with no measurable rise in claim denials.[2][5] In plain language, the tools helped providers capture more revenue per visit, and that extra revenue was not clawed back by payers. A separate analysis from the Blue Cross Blue Shield Association found that in some hospitals, the rate of diagnosed postpartum anemia climbed from 4% to 12.3% while the actual transfusion rate barely moved, and estimated that this coding shift alone added $22 million in maternity spending.[2][6]
I want to be careful here, because this is not a story about fraud. The care is real and the codes may be defensible. The point is subtler and more important: AI did not make care cheaper. It made the documentation of care more complete, and more complete documentation, under the way we pay for things today, costs more.
It gets one layer more complicated. Providers are using AI to capture more, so payers are buying their own AI to push back. The scribe company Abridge recently partnered with Nvidia and Eli Lilly and is expanding into coding and real-time claims work, while payers adopt claims-review AI of their own; one report noted that the two sides' tools may simply end up fighting each other over reimbursement.[7] So AI is not lowering the cost of the dispute between providers and payers. It is automating it, and arguably speeding it up.
And the hero has a bill of its own
Set the arms race aside for a moment, and there is still a problem with the rescue story. AI is not the cost-saver it was sold as, because of how it is priced.
Most enterprise AI runs on tokens. A token is just a small chunk of text the software reads or writes, and you pay for every one of them. It behaves like a utility meter, not a flat monthly subscription. The more your people use the tool, the higher the bill climbs, and the bill is not always predictable. Becker's reported this month on health systems scrambling to get that spending under control.[8] Uber, outside healthcare, capped how much AI its employees could use after the company burned through its entire annual budget for AI agents in the first three months of the year. Google reported its token usage rose sevenfold in a single year. The chief digital officer at Care New England warned that consumption-based pricing will cripple an organization that turns these tools loose before its teams understand how the billing works.[8] Houston Methodist decided it will not build an AI agent at all unless that agent can show a return, and it shuts off the ones that cannot.[8] Oracle has begun pricing some of its clinical AI by patient throughput, which means the cost rises right alongside your volume.[8][9]
So here is the shape of it. AI is a variable cost that scales with how much you use it, not a one-time purchase you make and forget. The leaders who modeled it as a productivity savings line are going to be surprised by an operating expense that grows the more successful the rollout becomes.
Why this is so hard, and what I think it means
Put the two halves together and you get a clear picture, even if it is not a comfortable one. There is no villain to defeat, and there is no hero coming to save you. The cost trend is the product of rational behavior across the whole system, and the technology we hoped would reverse it has instead become one of its drivers, while quietly running up a bill of its own.
I know that sounds bleak. I do not think it is. I think it is clarifying, because it points you at the work that actually matters instead of the work that merely feels good.
Here is my point of view. Stop waiting for the blame to be settled, because it will not be, and stop waiting for AI to fix coding and denials on its own, because it will not do that either. The organizations that come through the next few years in good shape will be the ones that do three unglamorous things. They will understand precisely which of these structural forces touch their own revenue cycle, rather than arguing about the system in the abstract. They will govern AI like the variable cost it is, holding every tool to a return measured in real dollars, days in accounts receivable, and patient satisfaction, and switching off the ones that cannot prove it. And they will keep enough skilled human judgment in the loop to know when the machine is capturing genuine complexity and when it is simply inflating the bill.
My forecast for 2030 is straightforward. The gap will not be between the systems that adopted AI and the systems that did not. Almost everyone will have adopted it. The gap will be between the leaders who treated it as a savior and let the meter run, and the leaders who treated it as a powerful, expensive tool that needs a firm hand on it. The first group will spend the decade looking for someone to blame. The second group will spend it managing the actual machine.
Healthcare has always been a relationship business, and the revenue cycle is no different. Technology will carry an enormous amount of the load by 2030. But the judgment about where to point it, what to pay for it, and when to overrule it, that part is still ours. There is no villain, and there is no hero. There is only the work, and the people steady enough to do it.
Sources
Pifer, Rebecca. "Americans mostly blame insurers for rising healthcare costs, survey finds." Healthcare Dive, June 10, 2026. https://www.healthcaredive.com/news/americans-blame-insurers-rising-healthcare-costs-survey/822399/
PwC Health Research Institute. "Medical cost trend: Behind the numbers 2027." PwC, June 11, 2026. https://www.pwc.com/us/en/industries/health-industries/library/behind-the-numbers.html
U.S. Government Accountability Office. "Health Care Consolidation: Published Estimates of the Extent and Effects of Physician Consolidation." GAO, 2025. https://files.gao.gov/reports/GAO-25-107450/
Condon, Alan. "Healthcare faces 'watershed moment' with costs jumping 9% in 2027: 7 things to know." Becker's Hospital Review, June 11, 2026. https://www.beckershospitalreview.com/finance/healthcare-faces-watershed-moment-as-costs-to-jump-9-in-2027-7-things-to-know/
Holmgren, A. Jay; Fenton, Cynthia L.; Thombley, Robert. "Ambient Artificial Intelligence Scribes and Physician Financial Productivity." JAMA Network Open, January 9, 2026. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2843524
Wennberg, David, et al. "Rising Coding Intensity and Its Impact on Health Care Affordability." Blue Cross Blue Shield Association, March 2026. https://www.bcbs.com/dA/70bb93b3a9/fileAsset/Rising-Coding-Intensity-and-Its-Impact-on-Health-Care-Affordability.pdf
Olsen, Emily. "Abridge partners with Eli Lilly, Nvidia as AI scribe eyes expansion." Healthcare Dive, June 12, 2026. https://www.healthcaredive.com/news/abridge-nvidia-eli-lilly-investment-ai-platform-expansion/822783/
Dyrda, Laura. "Health systems race to rein in AI costs." Becker's Hospital Review, June 2026. https://www.beckershospitalreview.com/healthcare-information-technology/ai/health-systems-race-to-rein-in-ai-costs/
Bruce, Giles. "Oracle Health targets double-digit growth with new AI EHR." Becker's Hospital Review, June 10, 2026. https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/oracle-health-targets-double-digit-growth-with-new-ai-ehr/
