I’ve seen a lot in my fifteen years in Silicon Valley. I’ve built companies, sold them, and invested in over 200 startups, including some of the biggest names in AI like Anthropic, OpenAI, and Scale AI. You get a pretty good nose for what works and what doesn’t. But nothing teaches you more than a good, old-fashioned failure. Especially one that consumes five years of your life.
This isn’t another story about a rocketship success. This is about the five years I spent grinding on a healthcare AI project that ultimately went nowhere. It’s a story about the hard truths and unexpected lessons that you only learn when you’ve got skin in the game. Building a strong AI model isn’t enough—especially in healthcare. From my experience, succeeding here means navigating tough regulations and market realities that most founders overlook.
The "Revolutionary" Idea
It all started with what felt like a world-changing idea. We were going to build an AI to predict the onset of clinical depression and anxiety from speech and text patterns. The concept was simple and powerful. People talk to their phones, they text their friends, they write emails. What if we could analyze that communication—with full user consent, of course—to catch the subtle linguistic markers of mental distress before it became a full-blown crisis? We could build a passive, early-warning system for mental health. It felt important. It felt necessary.
We assembled a killer team of machine learning engineers and computational linguists. We raised a seed round from investors who saw the vision. The first year was a blur of activity. We were building, iterating, and making incredible progress on the core technology. Our model was getting good. Frighteningly good. It could pick up on patterns of pronoun use, sentiment shifts, and response latency that were highly correlated with diagnoses from licensed therapists in our training data. We thought we had cracked it. We thought the hardest part was over.
We were wrong.
Hitting the Wall of Reality
The moment we stepped out of the lab and into the real world of healthcare, we hit a brick wall. It wasn’t one big thing, but a thousand small cuts that slowly bled our startup dry.
First came the regulatory nightmare. We knew about HIPAA, obviously. We had lawyers, we had compliance checklists. But knowing about it and living it are two different things. The sheer cost and complexity of ensuring every single byte of data was handled in a compliant way was staggering. Every potential partner, from a small clinic to a large insurer, had a different set of security requirements, a different vendor assessment process, and a team of lawyers whose job was to say "no." A simple data-sharing agreement could take nine months to negotiate. Nine months! In startup time, that’s an eternity.
Then we ran into the doctors. We had this beautiful, sophisticated AI, but the clinicians we showed it to were deeply skeptical. "How does it work?" they’d ask. When we tried to explain the intricacies of our transformer model, their eyes would glaze over. It was a black box, and they weren’t comfortable with a black box making recommendations about their patients ’s mental health. And honestly, I couldn’t blame them.
The Data Desert
Every AI lives on data. And we were starving. We had our initial training dataset, but to make the model truly robust, we needed a constant stream of new, real-world data. And that was a battle on all fronts. Getting users to consent to have their personal communications analyzed, even for a good cause, is a huge ask. The privacy implications are enormous. We spent countless hours and a small fortune on lawyers to craft privacy policies that were both compliant and understandable. But the friction was just too high.
We tried partnering with healthcare systems to get access to anonymized patient data. That was even worse. The data was a mess—fragmented, inconsistent, and locked away in ancient electronic health record (EHR) systems that felt like they were designed in the Stone Age. The data we needed was there, somewhere, but getting it out was like trying to squeeze water from a rock.
Who Pays for It? The Business Model Conundrum
Even if we solved the technical and regulatory challenges, we had a fundamental business model problem. Who was our customer? Who would pay for this?
We initially thought insurers would be eager to pay for a tool that could reduce mental healthcare costs. They weren’t. The sales cycle was brutally long, and they wanted hard proof of ROI, which we couldn’t provide without a large-scale, multi-year study. A study that would cost millions of dollars we didn’t have.
We considered selling directly to consumers. But who wants to pay a subscription fee for an app that tells them they might be getting depressed? It’s a tough sell. We were providing a solution to a problem people didn’t know they had yet. And the people who did know they had a problem were often the ones who could least afford to pay for a new service.
We were stuck. We had a powerful piece of technology that nobody wanted to pay for. We were a solution in search of a problem that had a budget.
The Painful Pivot and the Final Bow
After three years of hitting our heads against the wall, we knew we had to pivot. We tried to repurpose our technology for other markets. We explored using it for customer service sentiment analysis, for sales coaching, even for detecting deception in online communications. Some of these ideas had potential, but our hearts weren’t in it. We had set out to solve a major problem in mental healthcare, and everything else felt like a step down.
The team started to burn out. The initial passion and excitement were replaced by cynicism and fatigue. We were running out of money, and our investors were losing faith. After five years, we finally made the painful decision to shut it down. We returned the remaining capital to our investors, open-sourced parts of our technology, and went our separate ways.
What I Learned from the Ashes
Failure is a brutal teacher, but it’s an effective one. Here are the biggest lessons I took away from those five years in the trenches:
1. Technology is the easy part. We had a brilliant team and a powerful AI model. But in healthcare, the technology is only 10% of the battle. The other 90% is navigating the complex web of regulations, stakeholders, and incentives.
2. Don’t be a solution in search of a problem. We were so in love with our technology that we didn’t spend enough time understanding the real-world problems of our customers. We built a beautiful hammer, but we couldn’t find the right nail.
3. The "black box" is a dealbreaker. In high-stakes fields like healthcare, you can’t expect people to trust an AI they don’t understand. Explainability and transparency aren’t nice-to-haves; they’re essential for adoption.
4. Follow the money. A startup is a business, not a research project. If you can’t figure out who your customer is and how you’re going to get paid, you don’t have a business. It’s that simple.
I don’t regret the five years I spent on that failed project. It was a painful and expensive education, but it taught me more than any success ever could. It made me a better entrepreneur, a better investor, and a more realistic optimist. And it’s a story I’ll continue to tell, because in a world obsessed with overnight success, we need to talk more about the failures. That’s where the real learning happens.
Frequently Asked Questions
What was the biggest challenge in this case?
Almost always, the biggest challenge is people and alignment, not technology or strategy. Getting the right team focused on the right problem is harder than any technical challenge I've encountered.
What would you do differently looking back?
I'd move faster on the things that were working and cut the things that weren't sooner. Most founders, myself included, hold onto failing strategies too long because of sunk cost. Speed of learning is everything.
Can these results be replicated?
The specific numbers will vary, but the underlying patterns and principles are transferable. The key is understanding the context behind the results, not just copying the tactics. Every company has unique constraints that shape what works.
How long did it take to see results?
Most meaningful business results take 3-6 months to materialize. Anyone promising overnight success is selling something. The companies in my portfolio that grew fastest were the ones that stayed patient and consistent.