TL;DR
- The AI industry is obsessed with orchestration and hand-wavy about observation. A typical healthcare practice needs around 120 handshakes between a patient walking in and money hitting the bank, and the outcome needle does not move until all of them are fixed.
- Mid-market owners have no CIO and no consultant telling them about data sovereignty, so nobody warns them that their AI castle is built on someone else's land.
- Fixing one wedge accelerates work to the next bottleneck. Accounts receivable does not come down, and the rest of the 120 steps are a roadmap item for the vendor rather than a problem for you.
- The owner who ends up selling to a rollup at a low multiple owns no data and can show no AI-led efficiency. That is the trap door.
I have been on the road for three weeks now. Pensacola was pleasant. Memphis was muggy. Buffalo was chill. But the mood in the small business is one of nervous anticipation. It's like welcoming back a long-lost sibling, not knowing if she will be a delight or a piece of work. Only that this sibling is AI and she will stay with the owner, warts and all.
Contrast this mood with headlines touting gazillion dollar funding for AI companies solving problems in the pesky corners of healthcare, retail, and what not. The promises are spirited if you ask a CEO of a Rust Belt healthcare practice for her opinion. Over a beer she'd admit that the arrogance scares her. I see it. Tech has always been insufferable in how it talked down to their customer base. But this time, it comes not just from the position of know-how but from the delusion of knowing the unknown.
Why we went the other way
We are a tech company but we took a different path two years ago. That path involves working with business owners closest to traditional industries—those who talk the language, and have climbed the same mountains. They are our forward-deployed sherpas.
Day after day, I found myself in rooms where I felt I could explain the tech better but was wise enough to know that I was not the main man. Business owners showed their battle scars to each other, discussed AI in understandable terms, and looked to me for reassurance that they were thinking through it correctly.
I have come to hate the AI narrative.
Orchestration is loud. Observation is silent.
We have it wrong. We are obsessed with orchestration. We are hand-wavy about observation. A typical healthcare practice requires 120 handshakes from the moment a patient walks in until the money hits the bank. Judith might be slow but she has been around for 10 years and lent money to the owner when everything went south. They are family. Amanda has been through every one of the last five acquisitions in the state and knows every tool, every process, and why things are done in a certain way.
Kramer, the CEO is actually not a CEO. He is a physician forced to run a business. He knows the bank balance but not the ROI for his EMR system. When the EMR raises $250M and claims it can now do Amanda's and Judith's jobs 10 times faster, he does not care because he can fire them; he cares because he can ask Amanda to go integrate one more practice instead of drowning in claim-statusing.
The castle is on someone else's land
What they don't get is that their AI castle is built on someone else's land. Their data is the fodder to someone else's ambition. There is no CIO to advice. There is no Mckinsey consultant that talks to them about data soverignty. No one told them that they are simply accelerating AI to the next bottleneck and that their accounts receivables will not be coming down anytime soon.
No one comes down from the coast to observe what is done, what adds to complexity, why some jobs take the time they take. Each AI company listens to their masters' voice - focus, wedge, land and expand. In the meantime, the outcome needle does not move until all 120 steps are fixed. That's a 'later' problem for AI companies. In product-speak, those are things on the roadmap.
Meanwhile, the practice sells to a predatory PE rollup at 4x EBITDA. They own no data and show no AI-led efficiency. They are in a building with patients walking in.
What the owner actually wants
I don't care much for SaaS, or Service as Software or Forward Deployed Engineers. The owner does not. She does not care about outcome pricing. She wants a 101 on profit and how AI will improve it. She wants to know if she can manage one more location in spite of the staffing crunch she has. She has licensed six tools and all of them claim to be the latest AI. She knows Chad (that's what one of them called ChatGPT). All promises have been empty so far. ROI is a conversation that bores AI companies. They murmur the word 'outcomes' thrice in a call but there is no lamp to rub. Meanwhile they price by the seat and sell more AI module licenses.
What we do instead
We don't sell licenses. We don't sell hours. We sell faster reconciliation of billing to cash and take a portion for ourselves. We promise AI but only after we set up their own data lake, learn from it and deploy models trained on the patina of their evolution. We are making them a data company: we build AI models uniquely trained on their data, and we build or buy tools that align with their data sovereignty goals.
We are transforming a healthcare practice into a data-rich, context-coded, AI-ready business that can do better than 7x EBITDA of exit multiple – all without surrendering its fate to an AI platform that holds their context hostage. The trade you are being asked to make when you hand that data over is the subject of Don't shortsell your know-how to AI vendors.
If you are a tech company that truly cares about outcomes and you sell licenses, you are also selling a lie. It may not matter much when large corporations trade with each other. But in the small and mid-market world, the cost of that lie is someone's life's work.
Where Moative fits
Moative works as an AI operating partner with mid-market companies, and this essay is close to the reason the firm is structured the way it is. We take a share of the improvement rather than a license fee, which only works if we fix the whole path to cash rather than one wedge of it, and it only works if the data and the models end up belonging to the operator. In healthcare revenue cycle management that means mapping the full sequence from patient arrival to payment before automating any part of it, and in claims processing it means the denial logic stays yours.
Frequently asked questions
What is data sovereignty for a small business?
It means the operational data your business generates, and the models trained on it, remain under your control rather than becoming an asset on a vendor's balance sheet. For a mid-market operator with no CIO, it is usually the difference between building a company that is worth something on exit and renting one.
Why doesn't AI reduce accounts receivable?
Because a practice has roughly 120 steps between a patient arriving and money landing, and automating one of them moves work faster into the next bottleneck rather than out of the system. Days in AR respond to the whole path, not to the fastest segment of it.
What is wrong with land-and-expand AI vendors?
Nothing, as a business model for the vendor. The problem for the buyer is that the wedge is chosen for how easily it can be sold, not for where the constraint sits, and the remaining steps stay on a roadmap you do not control.
Does AI improve a company's exit multiple?
It can, but not through licenses. What an acquirer can price is owned data, demonstrable efficiency and models that transfer with the business. A practice that has spent three years feeding a vendor's model has none of those and looks, at diligence, like a building with patients in it.
Should a physician-owner buy AI tools or build capability?
Neither, framed that way. The useful first step is observation: work out where the 120 steps actually lose time and money, and which of them a model could help with once the data is shaped. Buying tools before that produces six licenses and no movement in profit.
What does an AI operating partner do differently from an AI vendor?
A vendor sells software and is paid whether or not the number moves. An operating partner is paid from the improvement, which forces attention onto the whole process and onto the data staying with the operator, because neither party gets paid otherwise.
