Five years. Over 1,800 days. That’s how long I spent trying to build the future of medical diagnostics. We had everything: a brilliant team of PhDs, millions in venture capital from top-tier funds, and an AI model that could spot anomalies in radiological scans with stunning accuracy. On paper, we were a rocketship. In reality, we were a slow-motion train wreck.
We failed. And it was the most important failure of my career.
I see a lot of founders today diving headfirst into AI, especially in healthcare, and I see them making the exact same mistakes I did. They’re mesmerized by the tech, by the sheer power of large language models and convolutional neural networks. They think that if they just build a better algorithm, the world will beat a path to their door.
I’m here to tell you that’s a fantasy. Building a great AI model is the easy part. It’s table stakes. Here's the uncomfortable truth about what it really takes to succeed in the brutal, regulated world of healthcare AI.
The Seductive Allure of the Algorithm
Our idea was born out of a genuine need. A close family member had a delayed diagnosis that could have been caught earlier with a more thorough review of their scans. I saw the potential for AI to act as a tireless, hyper-vigilant assistant for radiologists, catching the subtle patterns a human eye might miss after a long shift. We called the project “Aperture Dx.”
We spent the first two years purely on the tech. We assembled a massive, proprietary dataset of anonymized scans. We architected a novel neural network that achieved 98.7% accuracy in our lab environment. We published a paper that got some buzz. We raised a Series A from investors who saw the term sheet as a ticket to the next big thing in MedTech. We were convinced the hardest part was over.
I remember the day we got the model’s performance to beat the published benchmarks. We celebrated with champagne in the office. We felt like we had cracked the code. We hadn’t. We had just unlocked the door to a much harder set of problems.
The Brutal Truth #1: Your Model's IQ Doesn't Matter in a Clinical Setting
Our 98.7% accuracy? It meant nothing in the real world. The moment we moved from our clean, curated dataset to the messy reality of a hospital’s imaging archive, our performance cratered. The scans came from a dozen different machine vendors, each with its own quirks and artifacts. The patient data was inconsistent. The clinical context was missing.
We spent another year and a half just trying to get our model to be consistently “good enough” across this chaotic data. This is the unglamorous, back-breaking work of enterprise AI that no one talks about. It’s not about fancy algorithms; it’s about data pipelines, normalization, and building robust systems that can handle the garbage-in, garbage-out problem at scale.
The lesson: Stop obsessing over a percentage point of accuracy in a lab. Instead, obsess over workflow. How does your tool actually fit into the day of a radiologist who has to read 100 scans before lunch? If you make their life even 5% harder, they will never use it. Integration isn't a feature; it's the entire product.
The Brutal Truth #2: The Hospital Sales Cycle Is Where Startups Go to Die
Once we had a product that worked reasonably well in a pilot setting, we hit the real wall: selling it. We were a team of engineers and data scientists. We thought a great product would sell itself. We were naive.
Selling to a hospital isn’t like selling SaaS software. You’re not dealing with a single buyer. You have to convince the radiologist, the department head, the IT security team, the procurement office, the legal department, and the C-suite. Each one has a different set of concerns and the power to say no.
Our sales cycle for a single hospital was, on average, 18 months. That’s 18 months of demos, meetings, security reviews, and contract negotiations just to get a pilot project. For a startup burning through venture capital, that is an eternity. We burned through our entire Series A just trying to get our first three hospital contracts signed.
The lesson: You are not a tech company; you are an enterprise sales company with a tech product. Hire a seasoned healthcare sales leader before you write a single line of code. Understand the reimbursement landscape. Who pays for this? Is it a line item the hospital can budget for, or are you asking them to create a new one? If you don’t have a clear answer to the money question, you don’t have a business.
The Brutal Truth #3: You're Not Replacing Doctors. You're Augmenting Them.
One of our biggest early mistakes was our messaging. We talked about “AI-powered diagnostics” and how our tool could “reduce errors.” The doctors heard: “We think a machine can do your job better than you.”
That was the kiss of death. Doctors are highly trained experts who have spent a decade or more honing their skills. The last thing they want is some tech startup from Silicon Valley telling them how to do their job. We created an adversarial relationship from day one.
It took us a painful year to realize our mistake and pivot our entire approach. We stopped talking about accuracy and started talking about workflow. We positioned our tool not as a second opinion, but as a productivity tool. It could pre-screen scans, flag the most urgent cases for immediate review, and auto-populate reports with standard measurements. We weren’t replacing their expertise; we were saving them from the tedious, repetitive parts of their job so they could focus on the complex cases where their skills truly mattered.
The lesson: Frame your product as a tool that empowers clinicians, not one that threatens them. The goal is augmentation, not automation. Build something that makes a good doctor even better. That’s a product they will champion.
What I Learned From the Ashes
Aperture Dx never made it. We ran out of money chasing those long sales cycles and trying to perfect the tech in a vacuum. The acquisition offers were for the team, not the product, and we eventually wound things down. It was a gut-wrenching experience.
But the lessons were invaluable. They’ve shaped every investment I’ve made since, especially in the AI space. When a founder pitches me an AI company now, I don’t ask about their model’s F1 score. I ask them:
- Who is your sales leader and what’s their track record in this specific industry?
- Show me your workflow diagram. Where exactly does your tool fit, and whose job does it make 10x easier?
- What is your reimbursement strategy? Who cuts the check and why?
Most founders can’t answer these questions. They’re still stuck in the lab, tweaking their algorithms, celebrating that extra point of accuracy. They’re still making the same mistakes I did.
Don’t be one of them. The world of healthcare AI is littered with the ghosts of brilliant models that never saw a real patient. Success isn’t about having the smartest AI. It’s about having the smartest go-to-market strategy and a deep, humble respect for the complexity of the system you’re trying to change. That’s the brutal truth.
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.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
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.
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.