Essay
What commercializing AI in medicine taught me about selling it anywhere · Juan Vegarra
The hardest customer for AI is not the skeptic. It is the professional who is not allowed to believe you.
I have spent thirty five years selling new technology into industries I did not come from. Mining. Enterprise software. Cybersecurity. Today, medical devices. The pattern I am about to describe showed up in every one of them. But medicine is where it stops being a pattern and becomes a wall, and the wall has a name. The evidence bar is clinical.
In consumer AI, a demo closes the sale. The product does something remarkable in front of you, and you buy it, because the cost of being wrong is a bad afternoon. In medicine, a demo does not close anything. A demo opens a conversation that will run for years, through peer review, through reproducibility, through regulators, through committees whose entire job is to not be impressed. And that is correct. The clinician across the table is not being difficult. She is accountable in a way no software buyer has ever been. Her license, her patients, and her judgment are all on the line, and no demo on earth is allowed to substitute for evidence.
Most AI companies experience this as friction. I have come to see it as the single most useful teacher in commercial strategy. Because here is the uncomfortable truth: every AI market is quietly becoming a clinical market. The buyers are just at different stages of finding out.
For the past few years, AI has been sold on astonishment. The model does something that looks like magic, the room gasps, the pilot gets signed. That worked while the cost of being wrong stayed low and the novelty stayed high. Both are ending. CFOs are now asking what the pilot returned. Boards are asking who is accountable when the model is wrong. Insurers, auditors, and regulators are arriving in industry after industry, and every one of them asks the same clinical question: show me the evidence, not the demo.
Companies that learned to sell in medicine already know how to answer. Companies that learned to sell on astonishment are about to take a very expensive course.
First: sell the outcome, and let the outcome carry the burden of proof. A clinician does not buy an algorithm. She buys a better morning for a specific patient, supported by data she can defend to a colleague who is trying to find the hole. That discipline transfers everywhere. If your AI product cannot state its outcome in one sentence, with the evidence standing behind the sentence, you do not have a product yet. You have a capability in search of one.
Second: the model is not the moat. Models are becoming abundant. Whatever model you are proud of today, a comparable one will be rentable next year. What stays scarce is the instrument that generates the data, owned rather than rented, sitting at the point where real information is created and a real decision is made. The companies that will matter in five years are the ones building the flywheel: every use of the product improves the product, on data nobody else can buy. If your data arrives through a license, you have a head start. If it arrives through an instrument you own, you have a business.
Third: augment the professional. Never audition to replace her. The fastest way to lose a clinical buyer is to imply the machine will do her job. Not because she is defensive, but because she knows something the pitch deck does not: accountability cannot be delegated to a model. The AI that wins is the AI that makes the accountable human better, faster, and more confident, and says so plainly. This is not a messaging choice. It is a truth about how regulated professions adopt anything, and every profession is more regulated than your demo assumes.
One more lesson, learned expensively outside of medicine. Years ago I built a cybersecurity practice whose product was right and whose talent was eager, and it still moved slower than it should have, because the buyers already trusted someone else, and that someone saw us as competition rather than complement. The technology was never the problem. The channel was. AI companies are repeating this mistake at scale right now, selling around the consultants, integrators, and advisors their buyers already trust, and wondering why the pilots never convert. In medicine we learn this early, because you do not sell around the physician. You earn your way into the room she already trusts. Everywhere else, the same rule applies. It is just written in smaller letters.
If you sit on a board watching AI proposals arrive, borrow the clinical bar before your market forces it on you. Ask four questions. What is the outcome, in one sentence, and where is the evidence? Who owns the instrument that generates the data, and does every use improve the product? Who is the accountable professional, and does this make her better or audition to replace her? And whose trust does the buyer already rent, and are we inside that channel or fighting it?
Companies with good answers are rare. They are also recognizable in about twenty minutes, which is roughly nineteen more than a demo needs, and a full two years less than a bad pilot takes to fail.
The evidence bar is clinical now. Medicine just got there first.
Juan Vegarra is Chief Revenue Officer at VerAvanti and the author of An Outsider's Playbook (forthcoming). The views here are his own. More from the Notebook · Continue the conversation