Reimbursement Is the Real Product Strategy for Clinical AI

I am going to say the unglamorous thing that decides whether your clinical AI lives or dies, and it is not the model. It is reimbursement. STAT reported this month that a record number of AI-based medical devices qualified for Medicare add-on payments in 2026, the temporary sweeteners that help hospitals say yes to expensive new tech. Founders love to talk about accuracy and outcomes. The people with budgets are asking a colder question: who pays for this, how much, and for how long? If you cannot answer that on a napkin, you do not have a product. You have a science project with a pilot.

Accuracy gets you a pilot. Reimbursement gets you a business.

There is a gap I have watched swallow good products whole: the gap between a hospital trying your tool and a hospital keeping it. The trial is easy. Someone has innovation budget, your demo is sharp, everyone is excited. Then the budget cycle turns and the question becomes whether this thing pays for itself. STAT put the buyer's logic plainly: a technology is far more likely to stick if it can prove it drives down cost. Here is the shape of the money you are actually playing with.

Look at the third and fourth numbers, because that is where founders get surprised. The add-on is generous and it is temporary. It exists to get a new technology over the adoption hump, not to fund your company forever. Which means the clock starts the day you win it.

The temporary-payment trap

Think of the add-on as a bridge with a gap at the far end. It carries you across the first couple of years, and then it stops. If you spent that time celebrating the reimbursement win instead of building the evidence to earn permanent coverage, you walk straight off the edge. The whole job during the add-on window is to convert: gather the cost and outcome data that a permanent code, or a payer contract, will demand.

The overlap in that chart is the strategy. Your evidence build has to start the moment the add-on does and finish before it ends, so the permanent code is in hand when the bridge runs out. Teams that treat evidence as a launch afterthought discover, at year three, that they are asking hospitals to pay full freight for a tool that used to be subsidized. That conversation rarely goes well.

Design for the payer on day one

The mistake is treating reimbursement as a thing the finance team figures out later. It is a design input. Pick your path early, because each one asks something different of the product. CMS has been widening the early lanes too, with faster coverage for breakthrough devices through its RAPID pathway, but every one of these bridges still ends at the same place: you have to prove durable value. Here is the map I use.

Whichever row you pick, the product implication is the same: instrument cost and outcome from the very first deployment. If you cannot show a payer what you saved, you are asking them to take value on faith, and payers do not do faith.

The overuse trap, and why I would not celebrate too hard

One honest note, because it matters. Researchers in the same STAT piece warned that these add-on payments can incentivize overuse, nudging hospitals to run a device more than the patient actually needs because it pays to. If you build your model around volume that the incentive rewards rather than value the patient feels, you are building on sand. The subsidy will change, the code will get scrutinized, and a tool that only made sense when it was overused will not survive the audit. Design for the version of your product that is still worth paying for when nobody is subsidizing it.

So when I review a clinical-AI roadmap, the first slide I want is not the model card. It is the reimbursement plan: which path, what evidence, and how you cross the cliff. Get that right and accuracy becomes what it should be, the thing that makes your paid product good, rather than the thing you hoped would carry a product with no way to get paid.