I've seen a few bubbles in my day. After a decade in the Silicon Valley trenches, with a couple of exits and over 200 angel investments in companies from Anthropic to OpenAI, you get a feel for the hype cycles. And the noise around AI in radiology right now? It has that same frothy feeling. It’s a bubble, and I think it’s going to pop.
Too many founders are chasing a mirage. They think a model that’s 1% better at spotting nodules on a chest X-ray is their ticket to a billion-dollar company. They are dead wrong. I want to share the uncomfortable truth about what it really takes to win in the brutal, regulated world of healthcare AI.
The Big Mistake: A Great Model Is Worthless on Its Own
This reminds me of a mistake I made with one of my own startups. We built this beautiful, elegant piece of technology that solved a really hard problem. We were so proud. We were sure customers would be lining up. They weren't. We completely missed that our “brilliant” solution didn’t actually fit into their lives. It asked them to change how they worked, and the friction was just too high.
That’s exactly what’s happening in radiology AI. Founders are obsessed with model accuracy—the 98% vs. 99% debate. It’s a distraction. The algorithm is the easy part. The real monster is the workflow.
The Nightmare of Integration
A radiologist's workday is a painful juggle of different systems: PACS for images, RIS for information, and the EHR for patient records. These things are ancient, they don’t talk to each other, and they are where the work actually happens. A new AI tool, no matter how smart, is just another window, another login. It doesn’t integrate. It interrupts.
I once advised a startup that had a fantastic AI for spotting wrist fractures. The tech was legit. But they got zero traction. Why? A radiologist had to stop what they were doing, export an image, upload it to a separate website, wait, and then manually type the results back into the patient’s chart. It was a complete workflow disaster. They burned through their seed money and vanished.
The 2-Year Sales Cycle and the ROI Black Hole
Hospitals don't like new things. They are slow, risk-averse, and their purchasing process is a maze. It can easily take 18 to 24 months to close a deal. You have to convince the radiologists, the IT department, the lawyers, the compliance officers, and the CFO. Any one of them can kill the deal.
And if you survive that gauntlet, you face the final boss: the ROI question. How do you prove your AI, with its hefty annual subscription, actually saves the hospital money? Or that it improves patient outcomes enough to be worth it? That is an incredibly tough sell when most hospitals are already running on fumes financially.
The Real Goldmine: Boring Automation
So if diagnostic AI is a minefield, where are the real opportunities? The answer isn’t sexy, but it’s where fortunes will be made: boring automation.
The American healthcare system is choking on administrative tasks. It’s a mountain of paperwork, manual data entry, and wasted effort. This is where AI can make a huge, immediate difference. It’s not the stuff that gets you on the cover of Wired, but it’s the stuff that builds real, profitable businesses.
Just look at the problems:
- Prior Authorization: A soul-crushing process for everyone involved. An AI that could just handle the endless phone calls and faxes with insurance companies would be a hero to every doctor's office in the country.
- Medical Coding & Billing: This is a multi-billion dollar industry built on people manually translating doctor's notes into codes. It’s slow and full of errors. AI can do it better, faster, and cheaper.
- Finding Patients for Clinical Trials: Recruiting for trials is a huge bottleneck in developing new medicines. It's a manual chart review process that can take months. AI can scan millions of records and find the right patients in minutes.
These are just a few ideas. The point is, they solve a clear, expensive problem. The ROI isn't some fuzzy future promise; it's immediate and obvious. And the workflow is simpler—you’re replacing a spreadsheet or a fax machine, not trying to rewire a radiologist's brain.
My Own Pivot: Why I'm Pouring Money into AI for Mental Health
My own investment thesis has changed. I'm less excited about moonshot diagnostic tools and more interested in AI that attacks the system's inefficiency and access problems. It's why one of my recent angel investments is in the mental health space.
They aren't trying to build an AI therapist—that’s a clinical and ethical minefield I wouldn’t touch. They’re building the plumbing for therapy practices. They automate scheduling, billing, and insurance claims for therapists. It’s the “boring” backend work that burns therapists out and limits how many patients they can see.
By fixing the workflow, they are making therapy more affordable and accessible. The ROI for the therapist is a no-brainer, and they don't have to spend two years convincing a hospital committee to buy it. It’s a perfect case of finding a real, painful problem and solving it with a practical tool.
Stop Chasing the Sexiest Problem
The pull to build a world-changing, life-saving AI is strong. It feels like the most important thing you could be doing. But your impact isn't measured by the elegance of your algorithm. It's measured by whether people actually use your product.
The AI radiology field is full of founders who fell in love with their solution before they understood the customer's problem. They’ve built a beautiful hammer, and now they’re running around looking for a nail. The real money, and the real impact, is in building a better screwdriver.
My advice to any founder wanting to break into healthcare is this: ignore the sexy problems. Find the most boring, repetitive, and expensive administrative task you can, and build an AI to automate it. That’s where you’ll find your billion-dollar company. And you might just help fix our broken healthcare system along the way.
Frequently Asked Questions
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.
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.