TL;DR
- AI products in verticals are built in four layers: the data foundation, the platform plumbing, the product, and the use case. Venture-funded vertical AI companies are funded to build the third layer first, because that is what proves a market.
- That order is good for the vendor and bad for the buyer. You fund the foundation work through your own operating costs and wait for the vendor's roadmap to reach the bottleneck that is actually costing you money.
- A systems integrator inverts the order. It assembles AI tools that already work, fills the gaps between them, and spends its own effort on the data foundation and the integrations.
- Whoever owns the data owns the intelligence. A practice that uses AI tools and a practice that is genuinely AI-first are two different businesses to an acquirer, even when they run the same software.
Last week I wrote about how AI companies in verticals treat their customer base as fuel rods to build their nuclear power. This week I am going to expand on it and make a case for working with a systems integrator (one who puts tools and processes together) instead of working directly with an AI company.
How an AI product is actually built
To understand why you are a feed into a hungry machine, you need to understand how AI products are built. There are foundation layers:
The foundation: Data and context of everything you do as an organization
The platform: A set of tools that are assembled to deliver a product (Example: Document extraction, voice calling with AI, routing of work to AI agents, observing their work, etc.). Call this the plumbing.
The product(s): An assembled set of platform capabilities in familiar interfaces fitting into your habit of how you solve a problem today or get work done. (Example: Prior authorization check for healthcare practices)
The use cases: The same product that does prior authorization for physical therapy will also be relevant for cardiology, but the payers might be different and so there may be some last mile work to make the product fit into the cardio world. Same product, different manifestation.
Why the funding decides the order
Now, let’s come back to the dynamics of how AI products in verticals are funded. They raise a seed round to build the first version of the product (not the platform, not the foundation but the minimum effort to prove the market need). The reason they do the minimum is not because it is good for you.They do it because the capital they raise comes with condition: derisk first, prove that this is real before you scale.
This approach is a zero sum game. The vendor wins and you lose.
The prior authorization example
The best agentic prior authorization is not automatically going to reduce accounts receivable outstanding. You may get faster approvals because the payer too is using AI to adjudicate. But this does not mean that denials will remain the same. If you optimize to win the prior authorization battle with the payer, but the patient encounter documentation deviates from it, there will be a clawback.
I deliberately took what looks like a quick and easy win to show how frustrating it could turn. There are areas where the migration of the bottleneck is even more acute. But your Ai vendor that is specialized in prior auth or has automated prior auth first before anything else is simply asking you to hold the bag, make the investment, live with elevated cost of AI products, with nothing to show for – until they get to the next problem and the one after and so on. The mechanics of that bottleneck migration are the subject of The AI Trap Door, which follows a single fixed step through the rest of the path to cash.
Sounds easy?
No. Here is where the venture dynamic comes to play. To remove bottlenecks across the workflow, you need to first understand the full system. Understanding a system means bringing all the data into one place.That means custom work that happens in parallel to building their point attacks. Foundation piling, platform blocks, and confidence-winning product/use case deliveries all have to happen in parallel.
Why an integrator is structurally different
Structurally, whether a VC-funded Ai company does this or a system integrator that assembles solutions does this, how does it matter? The challenge remains the same, right?
No.
The system integrator takes 5 of these AI tools that are easy ready and stitches them together. They fill the gaps between them on their own. Instead they focus on the data foundation and integrations with and between the tools.
Integrations and moving data between tools is an infra/platform layer problem that the Ai companies will punt until they have unequivocally proven to their investors that they can take on more. The once that cannot prove, fall wayside or linger as zombies. Your future will be strung to a Zombie or vanish overnight, unless you too bet on the winning horse.
In AI, the winning horse is not the loudest EMR. It is the one that plays nice with multiple systems, gets ownership of your own data, makes sure the right tool is doing the right job and ultimately signs up to deliver you a better EBITDA.
That is a system integrator.
The operational case for owning your data
Why is data foundation important? Why is integrating and delivering the right handshake between processes crucial?
There is an operational reason and a strategic one.
Operationally, the promised land of better cost control and patient experience is possible only when data and context freely moves in and out of tools, in and out of the minds of these AI agents. More importantly, your data is too strategic to be owned and wielded by a third party AI company that learns from it, at its own pace, over time, with risks of falling wayside. Your operating success cannot be tied to the risky business model they have signed up on. Your practice cannot be the sacrificial lamb at the alter of world dominance hallucinations of a sleepy EMR or newly-minted valley-based AI company that is 2-years into healthcare.
The strategic case
Strategically, one who owns the data, owns the intelligence. Your system-wide, operational intelligence and actions delivered by best of breed AI tools but controlled and orchestrated by your system preserves your value. That argument is set out at length in Don't shortsell your know-how to AI vendors.
How?
Are you a physical therapy practice that uses AI tools or are you an autonomous, self-run, AI-first practice that focuses all its time on care delivery? The former is a company that will eventually sell to Select or USPH for 4x EBITDA. The latter has options: Scale, aggregate more practices, become a better run chain or sell to a strategic partner at 10x or more.
Where Moative fits
Moative is setting the wheels in motion to be this system integration partner, by joining hands with healthcare veterans who have been in the shoes of small, and mid-sized practices. We do this instead of raising venture capital, because not everything has to be about us and our success should not come at the cost of the one who pays our bills. In healthcare revenue cycle management that means building your data foundation before automating anything, and in claims processing it means the denial logic and the learning stay with you.If you are not in healthcare but the essay speaks to you, reply back. Let’s learn from each other.
Frequently asked questions
What is a systems integrator in AI?
A systems integrator assembles AI tools that already work, builds the integrations between them, and owns the data foundation underneath. It differs from a vertical AI vendor in that it does not need you to buy a single product, and it is not funded on the condition that it proves one wedge before touching anything else.
Should I buy from an AI vendor or a systems integrator?
It depends on whether your problem is one step or a chain. A vendor is the right answer when a single, well-bounded task is genuinely the constraint. An integrator is the right answer when the number you care about, such as days in accounts receivable, responds to a chain of steps rather than to any one of them.
Why do vertical AI startups build the product before the platform?
Because seed capital is granted on the condition that market need is de-risked first. The fastest way to prove need is a narrow product with a visible before and after. The platform and the data foundation are slower, less demonstrable, and therefore later.
Who owns my data when I use an AI vendor?
Read the contract rather than the marketing. The questions that matter are whether the vendor can train on your data, whether the resulting model is yours, and whether you can leave with your records and your derived context intact. If the answer to the last one is no, the vendor owns the intelligence even if you own the raw rows.
Does automating prior authorization reduce accounts receivable?
Not on its own. Faster approvals move work downstream to whatever the next bottleneck is, and if encounter documentation drifts away from what was authorized, the result can be a clawback rather than a saving. The AR number responds to the full path from patient arrival to payment.
What happens to my practice if my AI vendor fails?
That depends entirely on where your data and your integrations live. If both sit inside the vendor, a failure costs you the tooling and the accumulated context. If the foundation is yours, a failure costs you one tool, which you replace.
