I’ve seen a lot of things in my decade-plus 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 and OpenAI. But I’ve never seen anything like the gold rush happening in healthcare AI right now. And frankly, I’m worried.
I see hospitals and health systems throwing billions at AI, hoping for a magic bullet. They’re buying shiny new models, expecting them to transform patient care overnight. But the hard truth is, most of them are making a massive, $1.2 billion mistake. That’s not a made-up number. It’s the amount of money wasted on failed AI projects in healthcare just last year, according to a recent report I read.
Why is this happening? Because most founders and hospital executives 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.
The Siren Song of the Shiny New Model
I get it. The allure of a cutting-edge AI model is powerful. You see a demo, and it looks like magic. An algorithm that can detect cancer in a scan with 99% accuracy? Sign me up. An AI that can predict patient readmissions and save millions? I’ll take two.
But here’s the problem: a great model is just the beginning. It’s table stakes. The real work, the part that everyone seems to forget, is the integration. It’s the messy, unglamorous, and absolutely critical process of weaving that model into the complex fabric of a hospital’s existing workflow.
I learned this lesson the hard way with one of my own startups. We had built a phenomenal AI for radiology. It was more accurate than human radiologists in our lab tests. We were sure it was going to be a home run. We went to our first hospital pilot, full of confidence. And we fell flat on our face.
The radiologists hated it. Not because it wasn’t accurate, but because it was a pain to use. It didn’t integrate with their existing picture archiving and communication system (PACS). They had to open a separate window, log in to our software, and manually upload the scans. It was a clunky, time-consuming process that disrupted their entire workflow. The model was brilliant, but the product was a failure.
The $1.2 Billion Integration Gap
That experience taught me a crucial lesson: the value of an AI model is not in its accuracy, but in its adoption. And adoption is all about integration. You can have the most brilliant AI in the world, but if it doesn’t fit seamlessly into the way clinicians already work, it’s worthless.
This is the $1.2 billion mistake in a nutshell. Hospitals are buying models, not solutions. They’re focusing on the algorithm, not the workflow. And they’re paying the price in failed projects, wasted resources, and frustrated clinicians.
Here’s what the new data shows:
- 70% of healthcare AI projects fail to deliver any meaningful ROI. That’s a staggering number. And the primary reason for failure? Lack of integration.
- Clinician burnout is at an all-time high. And clunky, poorly designed AI tools are only making it worse. A recent study found that for every hour of patient care, physicians spend two hours on administrative tasks. We need AI that reduces that burden, not adds to it.
- The “last mile” problem is very real. Getting an AI model from the lab to the bedside is a long and arduous journey. It requires a deep understanding of clinical workflows, a commitment to user-centered design, and a willingness to iterate and adapt.
How to Avoid the Billion-Dollar Blunder: A Founder’s Playbook
So, how do we fix this? How do we bridge the integration gap and unlock the true potential of AI in healthcare? Here are the hard-won lessons from my own journey as a founder and investor:
1. Fall in love with the problem, not the solution.
This is a classic startup mantra, but it’s especially true in healthcare. Don’t start with a cool new model and then go looking for a problem to solve. Start with a real, painful problem that clinicians and patients are facing every day. Shadow them. Live in their shoes. Understand their workflows, their frustrations, and their unmet needs. Only then can you start to think about how AI can help.
2. Build a solution, not just a model.
Remember my radiology startup? We had a great model, but we didn’t have a solution. A solution is the whole package: the model, the user interface, the integrations, the training, and the support. It’s a seamless, end-to-end experience that makes the user’s life easier, not harder.
3. Think “workflow-native,” not “bolt-on.”
Your AI should feel like a natural extension of the user’s existing workflow, not a clunky add-on. This means deep integrations with the electronic health record (EHR), the PACS, and other clinical systems. It means a user interface that is intuitive, efficient, and even enjoyable to use. And it means a relentless focus on the user experience.
4. Co-create with your users.
Don’t build in a vacuum. Involve clinicians and other end-users in the design and development process from day one. Get their feedback early and often. Iterate based on their input. They are the experts in their own workflows, and their insights are invaluable.
5. Don’t underestimate the regulatory hurdles.
Healthcare is a highly regulated industry, and for good reason. Navigating the complexities of HIPAA, FDA clearance, and other regulations is a major challenge. But it’s not impossible. The key is to build a culture of compliance from the very beginning. Hire experts, invest in the right processes, and be prepared for a long and rigorous journey.
The Future is Integrated
The good news is, I’m starting to see a shift. A new generation of healthcare AI startups is emerging that gets it. They’re not just building models; they’re building solutions. They’re obsessed with workflow integration, user experience, and co-creation. And they’re the ones who are going to win.
I’m putting my money where my mouth is. I’m actively investing in companies that are taking this integrated approach. Companies that are solving real problems, building real solutions, and making a real difference in the lives of patients and clinicians.
The $1.2 billion mistake is a cautionary tale. But it’s also an opportunity. An opportunity to learn from our failures, to build better products, and to finally deliver on the promise of AI in healthcare. The future of healthcare AI is not about fancier models. It’s about deeper integration. And that’s a future I’m excited to be a part of.
Frequently Asked Questions
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.
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'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.
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.