Most founders think building a great AI model is enough. They're wrong. Here's the uncomfortable truth about what it really takes to succeed in the brutal, regulated world of healthcare AI.
I’ve seen it a hundred times. A brilliant team, a cutting-edge algorithm, a model with 99.5% accuracy on a curated dataset. They raise a seed round, burn through it building the “perfect” tech, and then slam headfirst into the brick wall of reality. The market doesn’t care about your AUC score. It cares about solving a real, painful problem for doctors and patients.
After two exits and over 200 angel investments in companies like Anthropic and Scale AI, I’ve developed a pretty good nose for what works. And what I’m seeing in AI-powered healthcare is a gold rush, but most people are digging in the wrong place. They’re looking for technical perfection. They should be looking for a path through the mud.
The Model Trap
Let me tell you a story. A few years back, we were building an AI diagnostic tool for radiology. The goal was to automatically detect early-stage lung cancer from CT scans. Our team of PhDs built a phenomenal model. On our test data, it was outperforming human radiologists. We thought we were on the verge of changing the world. We were not.
We took it to our first pilot hospital. The first scan we ran it on, it flagged a massive, obvious tumor as benign. A human radiologist would have spotted it from across the room. Our model, trained on pristine, perfectly labeled images, had never seen an artifact from a slightly miscalibrated scanner before. It just choked.
That was our first lesson: your model is only as good as the ugliest, messiest data you can feed it. The real world is not a Kaggle competition. It’s a chaotic mess of different machines, different protocols, and different operators. We spent the next six months not improving the model, but building a data pipeline that could handle the chaos. We had to get our hands dirty, manually cleaning and labeling tens of thousands of images from a dozen different hospital systems. It was brutal, unglamorous work. It was also the work that mattered.
The Data Nightmare
Everyone pays lip service to data, but few understand the sheer scale of the problem in healthcare. It’s not just about volume. It’s about a million other things:
- Privacy: HIPAA is not a suggestion. We had to build a completely new, on-premise deployment system because hospitals (rightfully) wouldn’t let patient data leave their walls. Cloud-native? Forget about it.
- Interoperability (or lack thereof): Getting data out of one hospital’s Picture Archiving and Communication System (PACS) and into another is a special kind of hell. We had to become experts in DICOM, HL7, and a dozen other arcane standards, just to get the data flowing.
- Labeling: You can’t just hire a bunch of people on Mechanical Turk to label medical images. You need board-certified radiologists, and they are expensive and busy. We ended up building our own internal labeling platform and paying top dollar for their time. It was the only way to get the quality we needed.
We probably spent three times as much on data infrastructure and operations as we did on the core AI research. And I’d do it again in a heartbeat.
The Sales Gauntlet
So you have a model that works on real-world data. You have a system that can be deployed in a hospital. Now comes the easy part, right? Just sell it.
Wrong. Selling to hospitals is a marathon, not a sprint. Our first big contract took 18 months to close. 18 months! We talked to everyone: radiologists, department heads, IT, legal, procurement, the C-suite. Each one had a different set of concerns, a different budget, and a different reason to say no.
Here’s what we learned:
- Find a champion: You need a doctor inside the system who believes in your product and is willing to fight for it. Our champion was the head of radiology at a major teaching hospital. She saw the potential and went to bat for us with the administration.
- Talk about workflow, not algorithms: Doctors don’t care about your convolutional neural network. They care about how your tool will save them time, reduce their workload, and help them make better decisions. We stopped talking about AI and started talking about a “smarter worklist” that prioritized the most urgent cases.
- Price for value, not cost: Don’t price your product based on how much it costs you to build. Price it based on how much value it creates for the hospital. We showed them how our tool could increase throughput, reduce errors, and improve patient outcomes. That’s a conversation they understand.
My Unfiltered Advice
Look, I’m not trying to scare you away from building a healthcare AI company. The opposite. I believe this is one of the most important and impactful areas to be working in right now. But you have to go in with your eyes open.
Stop chasing the perfect algorithm. Start by deeply understanding a clinical workflow and the real-world problems that doctors and nurses face every single day. Fall in love with the problem, not your solution.
Build a team that is as good at data engineering, regulatory affairs, and enterprise sales as it is at machine learning. And be prepared for a long, hard fight.
It’s a messy, brutal, and often frustrating journey. But if you can navigate it, you won’t just build a successful company. You’ll build something that actually matters. Something that saves lives. And that’s a prize worth fighting for.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.