The Case for AI Transformation in Healthcare
- Adi Tantravahi
- Aug 13
- 7 min read
Why the next five years will separate the organizations that adopted AI tools from the organizations that transformed with AI, and how you can make sure you're on the right side of that line.
The moment we're in
Every few decades, a technology arrives that doesn't just make work faster, but it changes what an organization is. We saw it with electricity, we saw it with the internet, and it's happening again with agentic AI.
But here's what most coverage of AI in healthcare misses: the opportunity isn't a better tool for the people you already have It's the ability to do more work and treat more patients without hiring more people, something every healthcare organization has been forced to accept as impossible for the last century.
Every organization in history has been built as a hierarchy, because hierarchies were the only way to move information: up from the front lines, decisions back down, coordination across teams. Layers of management exist not because anyone loves management layers, but because information doesn't move itself.
In healthcare, this has meant that administrators are bogged down with repetitive but high-value tasks that are ultimately a drag on the US economy.
We finally have technology that can act as the glue. Modern AI agents can carry information up and down an organization, coordinate work across functions, and execute the administrative tasks that hierarchy was built to manage. What's on the table isn't chatbots living in the corner of a workflow. It's the chance to change the operating structure of the enterprise itself.
Which raises a question worth sitting with before you evaluate another vendor. If the hierarchy you inherited was optional, what would you do differently? This post is what that means for healthcare, why the economics make this inevitable, and how you can take your organization to the promised land.
The economics of healthcare's broken production curve
Over the last fifty years, the number of physicians in the U.S grew roughly 200%, while the number of administrators grew roughly 4,000%, all while per-capita healthcare spending grew roughly 2,000% and productivity stayed flat.
The capacity to deliver care grew arithmetically while the administrative apparatus around it grew exponentially. Yet the production curve, the relationship between clinician capacity and the number of patients served, grew linearly because productivity stayed flat. Every additional patient has required additional people, and nothing has ever relieved that pressure.
Economists have a name for this trap: Baumol’s cost disease. Labor-intensive sectors like healthcare struggle to capture the productivity gains that transformed other industries like manufacturing and software, and as a result, their costs rise relentlessly relative to everything else. Healthcare administration is the purest case of this in the American economy. The work behind the scenes to deliver care, such as prior authorizations, billing, denials, eligibility, scheduling, coding, and follow-ups, is an extreme pattern of this coordination work done by people at a scale that compounds each year.
So far, software has had no impact on fixing this. EHRs digitized paperwork, and while they improved patient outcomes by making information readily available, they didn’t reduce the number of people required to push that paperwork. RPA also fell short of its promises, with brittle bots automating keystrokes while the coordination burden kept growing around the people it tried to automate.
Agentic AI is the first technology with the promise to bend this curve, because it doesn't just speed up a task inside the structure. It can replace the coordination function the structure exists to perform. That is the difference between a 5% efficiency gain and a different cost basis for running a healthcare organization.
“5% more efficient at a process that shouldn't exist is still 95% inefficient.”
The organizations that internalize this first won't just have a lower cost-to-collect. They will operate on a fundamentally different production curve than their peers, generating more revenue per FTE, getting cash in the door faster, and scaling administrative capacity without scaling headcount. These organizations could absorb growth without absorbing cost rather than having to choose between the two.
The path most organizations never explore
Every health system pursuing AI today is on one of two paths, whether they've named it or not.
Path one is tool adoption.
You buy AI features, give copilots to staff inside the existing hierarchy, and every person gets a little faster but the structure stays exactly the same. This is a safe path, and it produces comfortable results with single-digit efficiency gains that disappear into the noise of a budget cycle. Tool adoption automates down, taking the tasks at the bottom of the pyramid and shaving a few minutes off each one.
Path two is transformation.
Enterprises can redesign their operations around what AI can now own. Agents run entire workflows end to end, so humans move to the edge of the process and do the things only they can do: judgment calls, exceptions, building relationships, and delivering care. The coordination that required layers of management becomes intelligence built into the system, and the multiple levels of hierarchy collapse into two.
We think of this as automating up, not down. It's the difference between an AI initiative that shows up on a board slide and one that shows up in your financials.
The 5 levels of healthcare automation
Transformation is meant to be a climb, not a leap. We map every organization we work with against five levels. Finding yours is the first honest step.
Level 1 — Rules and robots.Leveraging macros, scripts, and RPA bots that follow explicit instructions on structured inputs. Useful, but brittle and expensive to maintain as every payer portal change breaks something. Most health systems have been here for a decade, and the ceiling is that rules can't handle the ambiguity that makes up most administrative work.
Level 2 — Assistive AI. leverage software with AI features built in. Copilots draft appeal letters, summarize accounts, and suggest codes. Individual staff get faster, but every workflow still routes through a human, and the org chart is untouched. This is where much of the industry sits today and where tool adoption stalls. The ceiling here is that you've made each seat more productive without needing more seats. Your capacity is still bounded by the number of people you have, which means the only way to grow is still to hire.
Level 3 — Autonomous workflows.Agents own discrete, well-defined workflows end to end, with humans reviewing by exception rather than by default. This is the first level where the production curve visibly moves, because work completes without a person in the loop. The ceiling is that these workflows are automated in silos, so the coordination between them is still entirely human.
Level 4 — Coordinated agent operations.Fleets of agents that work across functions, hand off to each other, escalate intelligently, and are managed the way you'd manage a team, with performance visibility, quality controls, and clear ownership. Supervisors oversee agent capacity the way they once oversaw human capacity, and the organization can restructure around this so there are fewer coordination roles, and more exception-and-judgment roles.
Level 5 — The agentic organization.Intelligence is the coordination layer, so information moves through the system without hierarchy moving it. Humans sit at the edge as clinicians, decision-makers, relationship owners, and administrative capacity scales with compute instead of headcount. The production curve is no longer linear. This is the destination, and no one in healthcare is fully there yet. The market leaders will be within the decade.
Two things are true about this pyramid. First, you can't skip levels. An organization that has never trusted an agent with a workflow can't leap to coordinated fleets. Second, the value is wildly non-linear. Levels 1 and 2 are measured in minutes saved. Levels 3 through 5 are measured in the shape of your labor cost structure.
Why you can't shortcut your way up the pyramid
Here's the uncomfortable truth about why so many AI initiatives in healthcare quietly stall at level 2: transformation is not a procurement decision.
Software vendors sell you a tool and a login, but the climb from level 2 to level 4 runs through the messiest parts of an enterprise across legacy systems, undocumented processes, payer idiosyncrasies, staff who have every reason to be skeptical, and workflows that exist only in the heads of the people running them. There isn’t a tool out there that can absorb these headwinds on its own.
Someone has to do the hard work of transformations: map the process, deploy against your actual systems, and earn trust one workflow and one team at a time. Then redesign the roles as agents take on more.
Trust is the hardest part of that work, and it should be. Handing a workflow to an agent in healthcare carries real exposure: audit risk, compliance obligations, and payer relationships that a single bad automated touch can damage. Leaders who slow down here aren't being change-averse, they're being responsible. But this is also why the climb can't be skipped. Agents earn autonomy the way a new employee does, by starting under full review and expanding scope only as accuracy holds.
The model that works, in healthcare and any other industry undergoing this shift, pairs the technology with the people deploying it. What you should be looking to buy isn’t software, but rather outcomes delivered by a team you hold accountable for making the transformation real inside your organization, on your systems, with your people. The technology matters enormously, but the technology alone cannot transform your organization.
Where the climb starts
Transformation carries the connotation of a multi-year odyssey. In practice it starts far smaller: one workflow with a clear financial signature, a measurable baseline, deployment against production systems with humans reviewing everything, and a deliberate redesign of the roles it touches.
Each pattern compounds. Each workflow makes the next one faster to deploy. Each accurate run earns the agent more autonomy. And somewhere along the way, the conversation shifts from "should we automate this task" to "what should this department look like."
What is at stake
American healthcare spends on the order of $1.5 trillion a year on administration to do coordination work that agentic AI can increasingly perform. That money isn't abstract, it's the gap between what care costs and what care should cost for every American, whether they’re a patient or an employer.
For your organization, the stake is more immediate. Bending this curve is the difference between an organization that can grow without growing its administrative cost base and one that will keep paying more every year for the same output, against competitors who no longer have to.
This transformation won't be accomplished by giving every administrator a copilot, but it will be accomplished by leaders willing to ask the rebuild question honestly: what would this organization look like if we rebuilt it today? And then to climb, level by level, until the answer and the reality converge.
The technology is ready. The economics are undeniable. The only variable left is which organizations move first.
Knowing your level is the easy part. Knowing where to start, and how to prove it worked, is where most organizations stall. We put together the framework we use to identify a first workflow, set a baseline, and measure the impact.
Comments