I once saw a pitch from a team of brilliant PhDs from Stanford. They had an AI model that could predict cardiac events with stunning accuracy. They thought they were going to change the world. They weren't. They ran out of money in 18 months.
Why? Because a great algorithm in a vacuum is worthless. Especially in healthcare.
I’ve spent more than a decade in Silicon Valley, building and selling companies, and now investing in the next generation of founders. I’ve backed over 200 companies, including some of the biggest names in AI like Anthropic and Scale AI. I’ve seen what it takes to win. And I can tell you, in the brutal, high-stakes world of healthcare AI, most founders are focused on the wrong things.
They’re obsessed with model accuracy. They’ll spend months trying to squeeze another half a percent out of their AUC curve. My team and I recently completed a massive analysis of over 10,000 clinical trials, and our own models can now predict the likelihood of a drug passing Phase 3 trials with about 87% accuracy. That number gets headlines. It sounds impressive.
But it’s not the most important part of the equation. Not even close.
The Uncomfortable Truth About Healthcare AI
Most founders think building a great AI model is enough. They're wrong. The model is just the ticket to the game. Winning the game requires something else entirely.
I learned this the hard way. One of my early angel investments was in a company—let’s call them "DiagnosTech"—that had a revolutionary AI for reading mammograms. Their model was better than human radiologists. Significantly so. We thought it was a slam dunk. The tech was incredible.
The company failed.
It failed because the founders, brilliant as they were, never truly understood the workflow of a radiology department. They built a beautiful, standalone software that required a radiologist to stop what they were doing, open a separate program, upload the images, and wait for the analysis. It was a disruption. It added friction to a system that is already overloaded.
No one used it. The radiologists went back to their old PACS systems, even if they were technically less accurate. The path of least resistance is a powerful force, especially in a hospital setting where every second counts.
This is the uncomfortable truth: The success of your healthcare AI company has less to do with your algorithm and more to do with your understanding of regulation, integration, and human behavior.
Lessons from the Trenches
After two successful exits of my own and looking at thousands of pitches, I’ve developed a pretty good filter for what works. It’s not about the buzzwords. It’s about a deep, almost obsessive, focus on the messy realities of the healthcare system.
Let's talk about drug discovery, the focus of our big analysis. That 87% accuracy figure is powerful. It means we can help pharma companies avoid spending hundreds of millions of dollars on drugs that are destined to fail. The potential is enormous. But an AI prediction by itself doesn't get a drug to market.
Here’s what actually moves the needle:
1. Stop Admiring the Problem. Solve a Tiny Piece of It.
Too many founders try to boil the ocean. They want to "revolutionize cancer treatment" or "end Alzheimer's disease." It’s a noble ambition, but it’s a terrible business strategy.
The companies that I see succeeding are the ones that pick a tiny, painful, and specific problem and solve it completely.
Instead of "AI for drug discovery," think smaller. Think, "An AI to predict the cardiotoxicity of kinase inhibitors for a specific patient sub-population with a rare genetic marker." Why? Because that’s a problem a specific pharma company has right now. It’s a problem that, if solved, has a clear, quantifiable ROI. You can save them $50 million on a single trial. That’s a sale you can make.
One of my portfolio companies is doing this in the radiology space. They aren't trying to replace radiologists. They built a tool that automates the incredibly tedious process of measuring and tracking tumor growth in CT scans for oncology trials. It’s a narrow, unsexy problem. But it saves radiologists hours per patient, reduces errors, and makes clinical trials run faster. They are now embedded in the workflow of three of the top five cancer centers in the country.
2. The FDA Is Not Your Enemy.
Founders are terrified of the FDA. They see it as a giant, bureaucratic wall designed to stop them. This is a huge mistake. The FDA’s mission is to ensure that drugs and devices are safe and effective. That’s your mission, too.
Treat the FDA as a partner, not an adversary. Engage with them early. Use their pre-submission programs. Show them your data. Be transparent about your model's limitations. The most successful companies I know have a regulatory strategy from day one.
They don't just build a model and then try to figure out how to get it approved. They design their entire product and data collection strategy with the FDA's requirements in mind. This isn't just about getting clearance; it's about building a better, safer product. The rigor of the FDA process forces you to validate your technology in ways that you otherwise wouldn't. It makes you stronger.
3. Build for the Nurse, Not the C-Suite.
The person who signs the check is rarely the person who uses the software every day. You might sell your vision to the hospital CEO, but if the nurses on the floor hate using your product, you are dead. I cannot state this enough.
Adoption is driven by the end-user. Your product has to make their life easier. Not in a theoretical, "this will improve overall hospital efficiency" way. In a tangible, "this saves me 10 clicks" or "this means I don't have to stay late doing paperwork" way.
Go shadow a nurse for a week. Shadow a lab technician. Shadow a clinical trial coordinator. Map out their entire workflow, down to the individual mouse clicks. Find the most annoying, repetitive, soul-crushing part of their day and build a tool that automates it. If you do that, they will become your champions. They will pull your product into the organization.
The 87% Is Just the Beginning
So what does our analysis of 10,000 clinical trials really tell us? It tells us that we are at an inflection point. The ability to use AI to predict outcomes with high accuracy is becoming a commodity. The real differentiator—the thing that will separate the next generation of unicorns from the graveyard of failed startups—is the mastery of the last mile.
It’s about integrating with messy legacy systems. It’s about navigating complex regulatory pathways. It’s about understanding the deeply human, and often irrational, behaviors of the people on the front lines of healthcare.
The next decade of AI in healthcare won't be defined by the elegance of the algorithm, but by the grit of the implementation. It’s not the sexiest part of the job, but it’s the only part that matters. Forget the hype. Go solve a real, painful, human-sized problem. That’s how you actually change the world.
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.