The Factory Floor: What Healthcare Gets Wrong, What AI Gets Right, and Where Vervint is Headed

It’s early November. I’m sitting in my mom’s hospital room. She’s fighting bladder cancer — one last push before hospice. I’ve been here for weeks, coming every night after work, logging many hours in this room, and I’ve had a lot of time to watch.

Nurses come in on a schedule. They badge into the laptop on the wall, pull up Epic, check the prescriptions and doctor’s orders. They scan her wrist. “Please state your name and date of birth.” Sometimes she knows the answer. Sometimes she doesn’t. They scan the medicines, run through their procedure, flash the badge, log out. The PCA comes in and does their work. The respiratory tech comes in and does theirs. Every person with a badge, every person with a process. If you ask a question, they write it down and route it through the system. A doctor appears once a day, on schedule, reviews the chart, talks through progress, leaves.

Three weeks later, on Thanksgiving Day, she passed away at home — her Christmas tree lit beside her.

Here’s what makes this hard to sit with: bladder cancer has a 95% five-year survival rate when caught early. From the first complaint of symptoms to an actual diagnosis took nearly a year, and honestly allot of fighting. The diagnosis came by accident — a sonogram during an unrelated ER visit I took her to for pain. By then, the cancer had spread to her lungs. She had seen doctors repeatedly for over a year. She’d even had an unnecessary surgery to remove her parathyroid. But no one ever ordered a comprehensive scan. A simple ultrasound, even a blood panel checking white cell counts, would have found it.

She didn’t fall through the cracks because anyone failed to care. Every nurse, every tech, every doctor was doing their job. They took those jobs because they want to help. She fell through the cracks because doing their job and diagnosing her cancer weren’t the same thing.

We work in IT consulting for the healthcare industry. We talk constantly about workflows — how they’re designed, how Epic implements them, how we transform them. But sitting in that hospital room, something clicked. We aren’t talking about patient care. We’re talking about a factory floor. And if what a patient needs is covered by a step in the factory’s process, they get good, efficient, effective care. If it isn’t, they’re on their own.

That realization is our starting point.

Healthcare Economics: A Market That Doesn’t Function Like One

Most markets work through a simple mechanism: a buyer and a seller negotiate over value. Competition rewards better products at lower prices. Over time, quality improves and costs fall. This is the engine behind nearly every industry that has gotten dramatically better and cheaper over the last century — technology, manufacturing, transportation.

Healthcare doesn’t work this way. And the reason isn’t complicated, even if the consequences are.

In a standard market, the buyer and the consumer are the same person. I want a phone, I buy a phone. I shop on price, reviews, and personal preference. The seller has to compete for me. In healthcare, the buyer is the insurance company. The consumer is the patient. They are not the same person, and they have fundamentally different interests.

Sellers — physician groups, pharmaceutical companies, device manufacturers provide the product, don’t negotiate with patients. They negotiate with insurers, through a hospital as a marketplace. Patients can’t shop on price because prices aren’t disclosed until after the service, and there isn’t a selection. They can’t choose based on quality because quality data isn’t available. And they can’t go without insurance because the uninsured price for basic care is unaffordable.

This creates what economists call a principal-agent problem, layered multiple times over. The plan member wants health outcomes. The plan sponsor wants cost control. The insurer wants to minimize claims payout. The physicians group wants higher compensation for its members. The hospital has to navigate all of it while keeping the lights on. By the time a patient walks into a room, four separate entities have already made decisions about what care they will and won’t receive.

The result is a system that spends more than any country on earth — $14,885 per person in 2024, more than double the average of comparable wealthy nations — and produces the worst outcomes in the developed world. Americans die four years earlier than citizens of peer nations. The United States ranks last among wealthy countries across nearly every dimension of health system performance. Higher spending correlates with *worse* outcomes, not better.

The hospital isn’t the villain here. It’s also trapped. It often must treat anyone who walks in the door regardless of ability to pay. It negotiates rates with over 900 different private payers, each with their own billing rules, coverage policies, and prior authorization requirements. Administrative costs consume 40% of total hospital expenditures — nearly $700 billion a year. Hospital administration alone costs twice as much per patient in the U.S. as in Canada.

A hospital cannot afford to organize itself around comprehensive diagnosis. It is paid for services rendered — individual transactions in a fee-for-service model. There is no economic reward for catching my mom’s bladder cancer on the first visit. There is, however, significant administrative liability for ordering too many tests that a payer might later deny. The factory floor is organized around what it gets paid to do. It gets paid to process encounters, not to achieve outcomes.

The Inflection Point

Healthcare has been financially stressed for years. It just got dramatically worse.

In March 2025, Congress passed a Continuing Resolution to avoid a government shutdown. For hospitals, it bought time by delaying cuts to Medicaid Disproportionate Share Hospital payments — funds that keep safety-net hospitals solvent. The cost was extending Medicare sequestration cuts through July 2032, clipping 2–6% of Medicare payments for every provider in the country for the next seven years.

Then came the One Big Beautiful Bill Act, signed on July 4, 2025. The Congressional Budget Office scored it at $1.02 trillion in Medicaid and CHIP cuts over ten years — the largest reduction in healthcare coverage funding in U.S. history. An estimated 7.6 to 10.5 million people will lose Medicaid coverage. Annual hospital revenue losses are projected at $24 billion per year. Safety-net hospitals face average operating margin reductions of 56%. Rural hospitals in states like Missouri and Montana face Medicaid funding cuts of 25–29%.

These cuts don’t drive patients away from hospitals. They drive uninsured patients *into* emergency departments, where care is unreimbursed. The Urban Institute projects an $18.9 billion *increase* in uncompensated care even as spending is “cut.” The system pays twice — once as delayed revenue and again as emergency care with no reimbursement.

By January 2026, hospital operating margins had already dropped to -0.6%. Expenses grew 5.4% year-over-year while revenue grew 3.9%. An estimated 750 hospitals are now at risk of closure. The major provisions of the OBBBA haven’t fully taken effect yet. The wave hits in 2027 and 2028.

Hospital leaders are not using words like “challenge” and “headwind.” They are using words like “structural” and “existential.” Kaufman Hall’s 2026 analysis put it plainly: “The current healthcare environment is not cyclical — it is structural.” The question for every health system in America right now is not whether to change. It’s whether they can change fast enough.

What AI Can Actually Change

This is where something new enters the picture.

AI is not new to healthcare. Administrative AI — billing automation, scheduling, coding assistance — has been growing for years. What’s new is something different: AI tools that replace clinical workflows rather than support administrative ones. Tools with FDA clearance to make diagnostic decisions. Tools that catch what the factory floor misses.

The regulatory landscape has shifted fast. The FDA cleared 1,451 AI-enabled medical devices and systems through the end of 2025. In 2025 alone, 295 were cleared — 76% in radiology. In 2026, Medicare introduced the first Category I billing codes for AI-assisted services. The economics of clinical AI just changed permanently: hospitals can now bill for it.

Aidoc’s CARE engine (FDA cleared January 2026) is the clearest example of what this looks like in practice. It’s a single AI platform covering 14 acute abdomen CT indications simultaneously — appendicitis, aortic dissection, bowel obstruction, liver and spleen injury, kidney stones, and more. Mean sensitivity 97%. Mean specificity 98%. It doesn’t wait for a radiologist to get to the scan in queue order. It analyzes every CT as it’s acquired, reorders the reading list to surface the most critical findings first, and flags them immediately. A deployment at Cedars-Sinai reduced CT turnaround time for brain bleeds from 53 minutes to 46 minutes. A Spanish hospital system using the earlier Aidoc hemorrhage detection tool saw 30-day mortality fall from 27.7% to 17.5% post-deployment.

Siemens Healthineers AI-Rad Companion extends this into multiple modalities. The Brain MR module (FDA cleared March 2026) segments 30 brain regions on MRI, measures volumes against a normative database, and flags deviations that may indicate Alzheimer’s or Parkinson’s. Previously, this required a subspecialty neurologist at an academic center. Now it runs on any standard MRI scanner and generates a structured report in minutes.

Digital Diagnostics LumineticsCore (formerly IDx-DR) is perhaps the most significant technology in the portfolio from a structural standpoint. It is the first FDA-authorized fully autonomous AI diagnostic system that requires no physician interpretation whatsoever. A medical assistant in a primary care office photographs the patient’s retina. The image uploads to a cloud server. The AI returns a binary decision: refer to ophthalmology, or return in 12 months. No doctor reviews the output. The AI detected diabetic retinopathy — the leading cause of preventable blindness in working-age adults — with 87.4% sensitivity and 89.5% specificity. The referral pathway from primary care to ophthalmologist collapsed into a single visit.

The economics are striking. A peer-reviewed 5-year ROI model for radiology AI showed 451% return on the workflow efficiency alone, rising to 791% when physician time savings are included. AI reduces radiologist workload by up to 53% — effectively doubling productive radiology capacity without a single additional hire. For a hospital system unable to fill open radiology positions and operating at negative margins, that’s not a technology investment. It’s a workforce problem solved by software.

The Mid-Market Gap

Large health systems with research arms like Mayo Clinic and Cedars-Sinai have the resources to partner directly with these AI companies. They have dedicated AI teams, exploratory research budgets, and the organizational capacity to run a 30% failure rate across pilot programs and absorb it. When an implementation fails at Mayo, they document the lesson and try again.

Mid-market health systems — those in the 200 to 2,000 bed range — do not have that buffer. The federal government’s own ONC data confirms the divide: multi-hospital system members adopt predictive AI at a rate of 86%. Independent community hospitals adopt it at 37%. That 49-point gap has not narrowed in two years.

More telling: even among hospitals reporting AI adoption, the vast majority are using administrative tools — scheduling, billing, coding. Clinical workflow replacement remains concentrated at the top. The mid-market is adopting AI that sits *adjacent* to existing workflows — ambient documentation tools that transcribe a physician visit, for instance, because the doctor still runs the visit and the AI just takes notes. That’s an adjacent change. A tool that replaces the radiologist’s reading queue, or that makes an autonomous diagnosis in a primary care office, is a workflow *replacement*. Many health systems follow academic leaders on workflow replacements. They don’t pioneer them.

This creates a defined and durable gap. These health systems want to adopt transformative AI. They are under financial pressure that makes the ROI argument compelling. They just can’t do it without help — not because they lack will, but because they lack the specific combination of skills needed to make it work.

The 80–95% failure rate for healthcare AI implementations is driven almost entirely by integration failure, not AI failure. Of 47 tracked AI integrations across 12 health systems, 41 failed — every one of them due to EHR integration barriers, data governance, or other implementation issues, not the AI tool itself.

That gap is not a failure of will or budget. It is a gap in a specific kind of expertise that most mid-market IT departments were never built to have in house: the work of turning a cleared AI tool into something that runs safely inside Epic, aligned to how a specific hospital actually delivers care.

Where We’re Headed

Vervint is focused on healthcare today. We migrate Epic to Azure. We run fully managed Epic environments. We provide Epic application support, infrastructure management, and custom application development, all within healthcare. That is where our relationships with these organizations started, and why we understand the ground these AI tools will need to run on.

The most important asset in healthcare AI deployment is not technical capability. It is trust. Trust is what lets a hospital’s leadership take a recommendation seriously. It is what lets clinical staff give a new workflow a real chance instead of working around it. That kind of trust is not for sale. It is earned over years of showing up reliably inside organizations that take patient care seriously.

We already have Epic and infrastructure depth. We already understand how health systems are organized, how clinical governance works, and how to navigate the intersection of IT and clinical leadership. That puts us inside the environment where AI deployment succeeds or fails, well before any AI tool enters the conversation.

We don’t run AI implementations today. What we’re doing, deliberately, is building toward that role: developing the Epic configuration expertise, developing AI vendor relationships, and the clinical workflow experience that turn a cleared AI tool into something a hospital can actually run safely. We would rather grow into this the right way than claim readiness we don’t have yet.

The goal has never been to become an AI company. It’s to become the partner that health systems trust to bring these tools in safely, the same way they’ve trusted us with Epic.

The Bigger Picture

AI is not going to fix American healthcare. The economic architecture is broken at a structural level — the principal-agent problem, the separation of buyer and consumer, the fee-for-service incentives, the administrative machinery that consumes 40 cents of every dollar before it reaches a patient. These are policy and economic problems. No software solves them.

But AI will disrupt healthcare in a way that goes far beyond workflow optimization. It will, for the first time, make comprehensive diagnosis accessible at the point of first contact — not as a luxury available to patients at major academic centers, but as a standard capability embedded in every general practice, every community hospital, every rural clinic connected to an EHR.

Picture this: A woman in her early 70s visits her primary care physician. She’s had some weakness the last few weeks. Intermittent lower abdominal discomfort. Blood in her urine once, which she mentioned almost in passing. Ambient listening documents the visit. AI audits the health record, flags the symptom combination against population-level risk models and surfaces an alert: this presentation, in this age group, with these specific symptoms, carries statistically elevated risk for urological malignancy. Evidence-based recommendation: order cystoscopy or CT urography within 14 days. The physician clicks through. The order is placed. One week later, a diagnosis comes back: Stage I bladder cancer. Highly treatable. Treatments scheduled.

This is not the future, the tools exist now. The FDA has cleared them. The AI works. More is coming quickly. What doesn’t yet exist, at scale, is the integration and implementation — the clinical workflow alignment, the Epic configuration, the change management, the governance — that makes it real for the community hospitals and regional health systems where most Americans actually receive their care. That is the gap we intend to help close.

AI will not fix everything broken about American healthcare. But it can close the kind of gap that took a year to find my mom’s cancer. That possibility is worth building toward, and we plan to be there as it happens.