We're hiring for the technology and firing for the technology in the same fiscal year
Twenty-one health systems have now named a chief artificial intelligence officer. Hackensack Meridian added the most recent one in August, hiring Memorial Sloan Kettering's chief data analytics officer away. Cedars-Sinai, Cleveland Clinic, Mayo, UCSF, and Mercy already had theirs. Twenty percent of enterprise healthcare CIOs cleared $600,000 in 2024, and IT salary budgets are up another three and a half percent this year.
In the same ten days, Indeed published its ranking of the healthcare jobs with the highest share of remote postings.
Coding specialist. Median pay $63,427. Utilization review nurse. $79,745. Care navigator. $63,319. Epic clinical analyst. $111,358.
Open any vendor deck about administrative automation and AI and you’ll find those same four jobs shown.
The top of the org chart is getting crowded
The chief AI officer role makes sense to me. Somebody senior should own governance when a tool can shape a coverage decision or a clinical note. My question is what the title is sitting on top of.
In August, the Center for Connected Medicine at UPMC and KLAS Research published interviews with twenty-seven health system leaders. Twenty-five have third-party AI deployed. Eleven have a platform or environment where they can test it. Sixteen called their AI strategy developing. One called it advanced. Eight of twenty-six are not measuring the value of their AI at all, are unsure, or are still planning how.
Twenty-one chief AI officers. One advanced AI strategy.
The titles are running about three years ahead of the capability, and a title has never once reviewed an output. A person does that, and she usually sits four levels down from the person with the title.
The bottom is getting thinner
Those four jobs landed on the remote list for good reasons. Coding, utilization review, care navigation, and Epic analysis are knowledge work that does not require standing in a building. Employers went remote to reach candidates for roles they could not fill locally.
People built their lives around that. They moved to towns with one hospital, or no hospital. They rearranged childcare. They took a $63,000 job specifically because it could be done from a kitchen table ninety minutes from the nearest health system.
The same qualities that made those jobs portable are the qualities that make them automatable: structured inputs, rule-heavy logic, high volume, measurable output. When McKinsey says half of administrative work can be automated and that agentic revenue cycle workflows can produce forty percent productivity gains, these are the jobs in the model. Not the chief AI officer's job.
The office is pulling back, too. Mayo ended remote-only job listings in May. Others have gone hybrid.
So the people in the most automatable roles are also the people with the least room to absorb the change. If your automation roadmap and your workforce plan are two separate documents owned by two different executives, you will not see this coming until it’s too late.
Somebody tried to staff the middle. He cannot get it funded.
Let's be honest. Most executives do not spend a minute thinking about who checks the machine's work. It feels like the plumbing. If nothing is backing up, it is not top of mind.
Surendra Khera at Cleveland Clinic Florida spent four years building that plumbing. He calls it a digital corridor: inboxologists, advanced practice providers, and virtual clinicians, all of them absorbing digital work before it reaches a physician. More than 20,000 visits handled. When a doctor takes vacation, the corridor clears seventy to seventy-five percent of the in-basket before she gets back. He expected to fight for recruits and filled an evening and weekend primary care pilot in two weeks.
Then he was asked about expanding it.
"The structure there is not well designed. The return on investments there are not clear. If they are doing work that has no tangible business model tied to it, other than offloading the clinicians in the physical space and retention, then I think it just makes it harder for them to understand the work and for us to expand that work from the business point of view."
He built the thing every AI strategy in this country quietly depends on. He proved people want to work in it. He cannot get it funded.
There is no billing code for the person who catches it.
We keep calling this a productivity problem
Here is my opinion: calling this a productivity problem is a dodge.
McKinsey put the numbers out in July. Labor productivity in clinical care organizations has fallen about one percent over twenty years. Across the broader U.S. services economy it rose more than fifty-five percent. Health systems now spend over $150 billion a year on IT. Hospital labor productivity peaked in 2001 and has been flat or falling ever since.
Productivity language puts it on the worker. The coder is slow. The nurse is inefficient. The registrar is not trying. Twenty years of flat output against a rising IT spend is not a story about people getting lazier. It is a story about an organization that never got aligned and bought software to cover the seams.
I have watched this happen for two decades. Patient access creates a problem that lands in mid-cycle. Mid-cycle creates a problem that lands in billing. Billing cannot fix upstream, so somebody buys a vendor to mop it up. Run that for fifteen years and you have three systems trying to fix the same encounter, and a productivity number that looks like your staff's fault.
It is not your staff's fault. It is THE structural problem, and AI on top of it just makes the bad handoff happen faster.
Watch where the money is coming from now
Summa Health in Akron converted to a for-profit organization last October, after a General Catalyst-backed firm bought it for $485 million. The incoming CEO starts in September. She is direct about the math: the community is aging faster than the workforce can grow, so technology has to carry it, back office first and clinicians later. Her colleague at the acquiring company has said plainly that Akron's exposure to Medicaid cuts runs steeper than diligence showed.
The University of Houston just opened a digital health institute with a venture studio attached. More than four million Texans live in rural areas. Nearly thirty percent of the state's counties have no hospital at all.
Neither of those is villainous. Capital is showing up where capital is needed.
But look at what is filling the gap. For most of my career, the counterweight to payer margin pressure and vendor overpromising was public, slow, and imperfect. Right now the fastest mover is private capital with a technology thesis, and the agency that used to ask the hard questions just told hospitals their administrative burden estimates were overstated because automation will absorb the work.
If the money and the regulator are both betting on the machine, who sponsors the person whose entire job is to say the machine got it wrong?
Nobody. That is the answer. Nobody sponsors her.
What should a CFO actually do about this?
Five things, in the order I would do them:
Count time, not titles. Pick three roles in your revenue cycle and log how a week actually gets spent. Sort every task into automation-ready, human-essential, or hybrid. You will learn very quickly which jobs are legacy constructs and which are load-bearing, and you will learn it before a vendor tells you.
Name the reviewer. For every AI tool running right now, write down the name of the human who reviews its output when it is wrong, and how long she has to do it. If you cannot fill in that blank, you do not have governance on that tool. You have a subscription.
Fund the middle on purpose. Khera's problem is a budget problem, not a strategy problem. If the only way to justify a review layer is a billing code, build an internal one. Track avoided rework, avoided denials, and avoided appeal cycles, and put a dollar figure on the catches. If the save is not visible, the seat never gets funded.
Protect the training ground. Your best coders learned to spot a bad code by working thousands of them. Your denial analysts learned the payer's tells by losing to them for a few years. If you automate every entry-level rep out of the workflow, you are not saving money. You are borrowing it from 2032 at a rate nobody quoted you. Medical schools are already restricting AI scribe access for trainees for exactly this reason.
Reprice the roles you keep. A person who reviews machine output, works the exceptions, and handles the patient conversation the machine cannot is doing more skilled work than her job description says, at a salary set back when the job was data entry. That gap is where your retention problem shows up first.
The seat nobody is paying for
By 2030 the revenue cycle runs on real-time adjudication, continuous eligibility, and prior authorization that clears at scheduling. I wrote a book about it and I still believe the forecast.
What I underestimated is how hard the middle would be to build.
Every plan in this industry ends with the same sentence. A human stays in the loop where judgment matters. Mine says it too, in print, in a book you can buy.
I have just stopped assuming anybody is paying for that human.
If this was useful
I wrote two books about exactly this. RCM Workforce Modernization covers role redesign, skill mapping, the 2030 talent map, and how to use federal funding to build the pipeline. Humanizing RCM covers the part that never fits on a dashboard. Both are on Amazon, along with RCM 2030: Strategy and Survival for Revenue Cycle Leaders.
If you have a vendor conversation on the calendar this fall, I built a free one-page worksheet with the three questions I would ask before signing anything. It is called The AI Vendor Test. [Download it here.]
I write RCM 2030 Weekly every Sunday for hospital CFOs and revenue cycle leaders. [Subscribe here.]
SOURCES
Becker's Hospital Review, "Which health systems have AI chiefs," August 2026
Becker's Hospital Review, "Hackensack Meridian Health names chief AI and data officer," August 2026
Becker's Hospital Review, "Pay for the 8 most remote-heavy healthcare jobs: Indeed," August 2026 (Indeed Hiring Lab, July 31, 2026)
Becker's Hospital Review, "CIO salaries, IT budgets climb: 10 things to know," August 2026 (Cherry Bekaert 2026 IT Salary Guide; WittKieffer)
Becker's Hospital Review, "Cleveland Clinic is building a new primary care workforce powered by AI, inboxologists and more," August 2026
Becker's Hospital Review, "Healthcare's productivity crisis won't be solved by AI alone: McKinsey," August 2026 (McKinsey, July 2, 2026)
Becker's Hospital Review, "Summa Health's next CEO on incorporating AI, navigating fiscal shoals," August 2026
Becker's Hospital Review, "U of Houston launches Institute for Digital Healthcare Transformation," August 2026
Center for Connected Medicine at UPMC and KLAS Research, "Validation and Trust: How Health Systems Are Testing and Governing Analytics and AI Solutions," August 2026
STAT, "Are AI scribes useful tools in medical education, or a crutch that imperils learning?" August 3, 2026
