Why are healthcare companies trapped in an AI ‘pilot purgatory’?

AI is seemingly everywhere in healthcare, but many life science companies appear to be stuck in what Arif Nathoo and Web Sun, co-founders of healthcare analytics AI company Komodo Health, deem “pilot purgatory.”

This means companies are running proof-of-concept pilots for various AI technologies but failing to take the next step to actually implementing the technology into day-to-day use, Nathoo and Sun explained to Healthcare Brew.

The root of the issue, according to Sun, is that many companies don’t have a clear, defined idea of what they want to use AI for.

“The challenge they’re experiencing is that most of these proofs of concept are designed to prove that the technology works and not to change the way a team or a business operates,” Sun said. “I think too many companies are in this proof-of-concept mode—or ‘pilot purgatory,’ as we dubbed it—to see if AI could help, to see if it works, but they’re not clear on what they’re trying to achieve.”

The healthcare industry isn’t alone. Preliminary findings from a 2025 MIT report found that 95% of companies got “zero return” from investing in AI pilots.

One of the issues, according to Sun, is the promise of AI “has been so dramatic” that some companies believe AI can do anything, so they use it broadly without thinking too much about specific use cases.

“Right now, they’re throwing meaningful budgets at AI initiatives, but they’re kind of scattershotting it across the enterprise,” he said. Total US healthcare AI spending hit $1.4 billion in 2025, according to an analysis from Menlo Ventures.

From life sciences to health systems. Hospitals, on the other hand, have largely gotten past the pilot stage, at least for administrative uses.

“Organizations have moved beyond pilots and are now strategically deploying solutions that directly impact provider burnout and the bottom line,” Trish Rivard, CEO of market intelligence firm Eliciting Insights, told Fierce Healthcare in March.

A 2025 survey published in the Journal of the American Medical Informatics Association found 100% of the 43 responding hospitals used AI for clinical documentation, with 53% reporting a “high degree of success.”

Where health systems haven’t made as much progress when it comes to AI is moving beyond administrative use cases to clinical ones, Marzyeh Ghassemi, an associate professor at MIT and lead of HealthyML, a group at the university that studies machine learning in health, told Healthcare Brew. AI has the potential to be used to speed up the diagnosis of chronic conditions like endometriosis, Ghassemi said, which currently takes an average of four to 11 years from symptom onset to diagnosis, according to the American College of Obstetricians & Gynecologists.

“If we can build AI systems that can do early detection of risk for endometriosis, then it’s easy to screen,” she said. “We can send people for screening much earlier and then have a whole generation of patients that have access to care much earlier.”

However, that’s not where most investment into AI is going, according to Ghassemi.

“It’s a little bit weird to look at where AI is being deployed and what kinds of things it’s being used for,” she said. “It’s really where there’s a lot of short-term profit to be had…and not through improving the fundamental capacities that clinicians have.”