This is a draft and may not be the final version.
7 Things I Learned Angel Investing in 20+ AI Healthcare Startups
I once lost $50,000 on a startup that had, without a doubt, the best damn algorithm I had ever seen. The team was stacked with PhDs from Stanford and MIT. Their tech could predict the progression of a rare neurological disease with stunning accuracy. On paper, they were unstoppable. They went bust in 18 months.
Why? Because not a single doctor would use their software. It was clunky, it didn’t fit their workflow, and it required them to open a separate application. The best model in the world, and it was defeated by a login screen.
That failure taught me more than any of my successes. I’ve been in Silicon Valley for a while now. I’ve built and sold two companies—you might have heard of RemoteTeam, which we sold to Gusto, or MovieLaLa, which went to Gfycat. Now I spend most of my time investing in the next generation of founders. I’ve made over 200 angel investments, a good chunk of them in the AI space, including some names you’d recognize like Anthropic and Scale AI. But my real passion lately has been healthcare.
I’ve put money into more than 20 AI healthcare startups, and I’ve seen it all. The brilliant successes, the spectacular flameouts. And I can tell you this: the rules are different here. What works for a SaaS company or a consumer app will get you killed in healthcare. 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.
1. The Model is the Easy Part
This is the first thing that trips up almost every technical founder. They spend a year perfecting their model, getting it to 99.8% accuracy, and they think they’re done. Look, I get it. As an engineer, that’s the fun part. But in healthcare, the model is just the ticket to the game. It’s table stakes.
Your competitors have smart people, too. They probably have access to similar datasets. That tiny edge in accuracy you’re so proud of is not a defensible moat. The real work, the brutal, unglamorous work, starts after the model is built. It’s about integration, workflow, and dealing with the messy reality of hospital IT systems that look like they were designed in 1998.
2. Data Isn't a Moat, Workflow Is
Every pitch I hear has a slide that says “data is our moat.” It’s almost never true. Sure, proprietary data is great if you can get it, but it’s the workflow that will make or break you. You have to build something that a doctor or a nurse can use without thinking, something that slots directly into the chaos of their day.
I invested in a company doing AI-powered radiology diagnostics. Their first product was a standalone app. It was beautiful. It was also a complete failure. Radiologists had to stop what they were doing in their main system (the PACS), open our app, find the patient file, look at the results, and then go back to the PACS to write their report. It added three minutes to every single scan. They were never going to do that. The company nearly died before they pivoted. They spent a year building a native integration that put their insights directly inside the PACS viewer. It was a painful, expensive process. But once they did, adoption exploded. They weren’t selling an AI model anymore; they were selling a faster, easier way to do the job.
3. Regulation is Your Friend (Seriously)
Most founders see the FDA and other regulatory bodies as giant monsters to be avoided at all costs. They’re terrified of the paperwork, the clinical trials, the endless delays. I see it differently. Regulation is a barrier to entry. It’s a giant wall that, once you climb it, keeps all the fast-following copycats out.
Yes, it’s a slog. It can cost millions of dollars and take years. But it forces a level of discipline and rigor that most startups lack. It makes you prove that your product is not just cool, but safe and effective. And once you have that FDA clearance, you’re in a very exclusive club. You can command higher prices, and you have a story that gives hospital administrators the confidence to sign a seven-figure check. Don’t run from the FDA. Use it.
4. Sell to the CFO, not the CMO
Doctors are your users, but they are not your buyers. This is a critical distinction. A doctor might love your tool, but she doesn’t control the budget. The person who does is the Chief Financial Officer or some other hospital administrator whose main job is to control costs.
You can’t sell them on fancy tech. You have to sell them on ROI. Hard, quantifiable, dollars-and-cents return on investment. How much time will your tool save? How many fewer mistakes will be made? How will it reduce the length of a hospital stay? You need to have a rock-solid answer to these questions. I passed on a company with a brilliant diagnostic tool because their only pitch was “it’s more accurate.” They couldn’t explain how that accuracy would translate into cost savings. They never got a single hospital contract.
5. "AI for X" is a Terrible Pitch
If the first words out of your mouth are “We’re using AI to…”, I’m already losing interest. I don’t care about your technique. I care about the problem you’re solving. Stop leading with the implementation detail and start with the pain point.
Instead of “We’re using a novel transformer architecture for drug discovery,” try “We can cut the pre-clinical phase of drug development in half, saving pharma companies hundreds of millions per drug.” See the difference? One is a tech demo, the other is a business proposition. If you want to learn more about what I look for in a pitch, you can read my post on how to pitch me.
6. The ‘Human-in-the-Loop’ Trap
Everyone starts with a “human-in-the-loop” model. It’s the only way to get going when your AI is still, frankly, a bit stupid. You have human experts who review the AI’s output, correct its mistakes, and generate the labeled data needed to make it smarter. It’s a necessary evil.
The problem is, many companies never escape it. They get stuck. Their costs stay high because they need a small army of experts on payroll. They’re not a scalable software company; they’re a tech-enabled services company. And those get much lower valuations. You need a clear, credible plan to reduce your reliance on the humans. What’s your path to 90% autonomy? 95%? If you don’t have one, you don’t have a venture-scale business.
7. Founder-Market Fit is Everything
This is maybe the most important lesson of all. Why are you the person to solve this incredibly hard problem in this incredibly difficult industry? What unique insight do you have?
I’ve seen too many smart outsiders try to barge into healthcare and fail because they don’t understand the culture, the incentives, the politics. The founders who succeed are the ones with a deep, personal, almost obsessive connection to the problem. They’re the surgeon who got fed up with outdated tools and decided to build his own. They’re the patient who nearly died because of a misdiagnosis and has now dedicated her life to fixing the system. They have a fire in their belly that no amount of money can buy. They have a right to win.
Investing in healthcare AI is not for the faint of heart. It’s a long, hard road, and the odds are against you. I’ve made my share of mistakes, and you can read about them in my biggest investment mistakes. But the potential impact is staggering. We’re on the cusp of fundamentally transforming how we diagnose, treat, and manage disease. If you’re a founder in this space, don’t just focus on the algorithm. Focus on the workflow, the business case, and the human problem you’re trying to solve. If you can do that, you might just build something that lasts.
Frequently Asked Questions
Which item on this list has the highest impact?
It depends on your stage and context, but in my experience, the items near the top of the list tend to have the broadest applicability. That said, sometimes the less obvious items create the biggest breakthroughs for specific situations.
Can I implement all of these at once?
I'd strongly recommend against it. Pick the 2-3 items that resonate most with your current situation and focus there. Trying to do everything simultaneously is a recipe for doing nothing well.
How do I know which items apply to my situation?
Start by honestly assessing where your biggest bottleneck is right now. The items that address that specific constraint will give you the highest return on your time and energy.