I’ve seen a few hype cycles in my day. I built and sold two companies here in Silicon Valley, and now I spend my time investing in the next generation of founders—over 180 of them so far, including some you’ve probably heard of like OpenAI and Anthropic.
And I’m telling you, the buzz around “AI for radiology” feels just like a bubble getting ready to pop.
Every week, I get a pitch that goes something like this: “We have a proprietary dataset of 100,000 chest X-rays and a world-class team of data scientists. Our model can detect pneumonia with 99.7% accuracy, better than a human radiologist!”
I pass on almost every single one.
It’s not because the tech isn’t impressive. It often is. It’s because the founders are solving the wrong problem. They think the key is building a better algorithm. It’s not. Most of these companies are going to fail, and I’d rather share the uncomfortable truth now than watch more founders waste years of their lives.
The Uncomfortable Truth About Healthcare AI
When we were building RemoteTeam, which Gusto later acquired, we weren’t just building software. We were changing how companies operated. We had to integrate with dozens of messy, legacy HR systems. It was a painful, brutal process. That experience taught me a lesson that’s a hundred times more critical in healthcare: the algorithm is the easy part.
Most AI radiology startups are doomed for reasons that have nothing to do with their model’s F1 score.
1. The Data Moat is a Mirage
Founders love to talk about their “proprietary data.” The reality is that hospital data is a nightmare. It’s a chaotic mess of different formats, stored in siloed, on-premise servers that look like they were built in the 90s. The data is often unlabeled, poorly annotated, and full of errors.
One of my portfolio companies spent a year and nearly a million dollars just trying to clean up and standardize a dataset from a single hospital system. They discovered the same patient had three different IDs, scans were mislabeled, and the notes from doctors were basically unusable.
Your unique dataset isn’t the moat you think it is. It’s a swamp. And while you’re wading through it, ten other startups are wading through their own swamps, and the big incumbents like GE, Siemens, and Philips have been draining those swamps for a decade.
2. Your Workflow is a Radiologist’s Nightmare
Imagine you’re a radiologist. You’re reading hundreds of scans a day, under immense pressure, with a dozen different windows open on your screen. Your workflow is a finely tuned machine built for speed and efficiency. You have seconds to make a call.
Now, here comes a startup with a new AI tool. To use it, you have to open another window, log in to another system, and wait for some cloud-based model to process the image. It might be 1% more accurate, but it just added 30 seconds to your workflow for a single scan.
It’s a non-starter. I saw a brilliant team with an FDA-cleared algorithm for detecting wrist fractures fail because of this. Their model was incredible, but it wasn’t integrated into the standard PACS (Picture Archiving and Communication System). Radiologists had to use a separate web portal. Adoption was zero. The company folded 18 months later.
If your product doesn’t seamlessly integrate into the existing workflow and make a doctor’s life visibly, immediately easier, it’s dead on arrival.
3. The Sales Cycle from Hell
Selling to hospitals isn’t like selling SaaS software. You can’t just run some Facebook ads and get sign-ups. The sales cycle is brutally long. We’re talking 18, 24, even 36 months from first contact to a signed deal.
You have to convince the radiologist, the department head, the IT department, the legal team, the procurement office, and the C-suite. Each one has their own budget, their own politics, and their own reasons to say no. The hospital’s IT team is terrified of a data breach, the lawyers are worried about liability, and the CFO wants to see a clear ROI before spending a dime.
Most startups will burn through their entire seed round before they close a single enterprise deal.
Where to Focus Instead: The “Boring” Problems
So if radiology AI is a bubble, where should ambitious founders be looking? My advice is simple: stop chasing the sexy diagnosis algorithms. The real opportunity is in solving the boring, unsexy, operational problems that plague healthcare.
Think about the machinery that runs around the doctor.
Clinical Trial Recruitment: It costs pharmaceutical companies billions and can take years to find the right patients for a clinical trial. An AI that can scan millions of electronic health records to identify eligible candidates in minutes? That’s not just a business, that’s a revolution. It’s a clear, multi-billion dollar problem.
Automating Administrative Work: Doctors and nurses spend up to a third of their day on paperwork, billing codes, and insurance claims. It’s a soul-crushing waste of time and a major driver of burnout. Build AI that automates this. Use large language models to listen to a doctor-patient conversation and automatically generate the clinical note, the billing codes, and the prescription order. That’s a product a hospital will actually buy because it saves them money and makes their doctors happier.
Drug Discovery: This is the high-risk, high-reward frontier. Instead of reading images, you’re analyzing genomic, proteomic, and chemical data to find new drug targets and predict compound efficacy. The science is incredibly complex, but the payoff is astronomical. A successful drug is worth billions. This is where my investments in companies like Scale AI and Hugging Face come into play—they are building the foundational infrastructure to process the massive datasets required for this kind of research.
Stop Chasing the Gold Rush
Look, I get the appeal of building an AI that can “beat the doctor.” It’s a compelling story. It’s a great headline. But it’s a terrible business strategy.
The history of technology is filled with stories of companies that had amazing tech but failed because they didn’t solve a real, painful business problem. The AI radiology space is shaping up to be another chapter in that book.
Don’t be another casualty. Stop chasing the sexy algorithm. Go find a boring, expensive, painful operational problem and solve it.
Solve the scheduling nightmare. Solve the billing mess. Solve the clinical trial bottleneck. Be the company that gives doctors back an hour a day.
That’s how you’ll build a real, enduring company in healthcare. That’s where you’ll find your exit.
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.
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.