Most founders think building a great AI model is enough. They’re dead wrong. Here’s the uncomfortable truth about what it really takes to succeed in the brutal, regulated world of healthcare AI.
I’ve been through the startup wringer a few times. Two exits, over 200 angel investments in companies like Anthropic and OpenAI. You'd think I'd have seen it all. But nothing—and I mean nothing—prepared me for the unique beast that is healthcare AI.
We had everything going for us. A team of brilliant PhDs from Stanford and Berkeley, a model that was outperforming benchmarks, and early buzz from investors. We were building a general-purpose diagnostic AI. Give us any medical scan—X-ray, MRI, CT—and our model could spot anomalies. We thought we were building the future of medicine. We were on top of the world.
And then we hit a wall. A big, fat, FDA-shaped wall.
The Generalist’s Curse
Our model was good at many things, but not great at any one thing. When we took it to hospital administrators, they’d ask, "Okay, so it can detect 50 different things with 90% accuracy. But our radiologists are already at 98% for the top 5 most critical conditions. Why would we switch?"
They had a point. We had built a powerful engine, but we had no idea what car to put it in. We were a solution in search of a problem. The tech was impressive, but the business was a mess. We spent months trying to force it into different clinical workflows. None of them stuck.
Our burn rate was terrifying. We had 18 months of runway, and in Silicon Valley, that feels like 18 minutes. We were six months in, with nothing to show for it but a fancy model and a deck full of vanity metrics. I remember one particularly brutal board meeting where one of our investors, a guy who’d made his fortune in enterprise SaaS, just laid it out: "I don't get it. You have the best tech, but you can't sell it. What are you missing?"
He was right. We were missing focus. We were trying to boil the ocean.
The Pivot That Saved Us
That night, I couldn't sleep. I was staring at the ceiling, thinking about all the people who had bet on us—our team, our investors, our families. I felt like I was letting them all down.
I started thinking about my portfolio. I’ve been lucky to back some incredible companies in the AI space. What did the successful ones have in common? They didn’t build general-purpose AI. They found a tiny, painful, and expensive problem and solved it better than anyone else.
That’s when it hit me. We needed to stop trying to be everything to everyone. We needed to find our niche.
I called an emergency all-hands meeting the next morning. I told the team we were ditching the generalist approach. We were going to find one specific, high-value problem in healthcare and own it. We spent the next week locked in a conference room, brainstorming and arguing. We talked to dozens of doctors, researchers, and pharma execs.
And then we found it. A tiny corner of the drug discovery world that was completely underserved by modern technology. It was a specific type of protein analysis for a rare class of diseases. The existing methods were slow, expensive, and manual. It was a $10 million problem that nobody was paying attention to.
It was perfect.
From Broad AI to Sharp Tool
We threw out 80% of our code. We retrained our model on a highly specific dataset. It was painful. It felt like we were going backward. But for the first time, we had a clear target.
Within three months, we had a prototype that was 10x faster and 5x cheaper than the existing solution. We took it to a mid-sized pharma company we’d been talking to. Their head of R&D was skeptical at first. He’d seen a hundred AI demos that promised the world and delivered nothing.
But when we showed him our results, his jaw dropped. He couldn’t believe it. We signed a pilot contract on the spot.
That pilot turned into a multi-year, seven-figure deal. And that deal gave us the validation we needed to raise our next round, this time at a valuation that made our seed investors very happy.
What I Learned From Almost Dying
Looking back, that near-death experience was the best thing that ever happened to our company. It forced us to confront the hard truths about building a business, not just a piece of technology.
Here’s what I learned:
- Focus is everything. In the world of AI, it’s easy to get seduced by the power of the technology. But a powerful model is useless without a specific, valuable application. Find a niche, own it, and then expand from there.
- The market doesn’t care about your tech. It cares about its problems. Stop talking about your model’s accuracy and start talking about how you can save your customers time, money, or headaches.
- Regulation is not a bug, it’s a feature. The healthcare industry is a maze of regulations for a reason. If you can navigate that maze, you have a powerful moat that will protect you from competitors. Don’t fight it; embrace it.
Building an AI startup in healthcare is not for the faint of heart. It’s a long, hard slog. But if you can find that perfect intersection of a powerful technology and a painful problem, you can build something truly transformative. And maybe, just maybe, you’ll even make it out alive.
I'm not saying it's easy. It's not. But it's worth it. The feeling of seeing your technology actually make a difference in the world is something that no amount of money can buy. So if you're out there building something in the healthcare space, don't give up. Find your niche, solve a real problem, and you might just change the world.
Frequently Asked Questions
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.
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.