The Problem with Healthcare AI Isn’t the Tech – It’s Venture Capital Timelines

If healthcare AI has hit the trough of disillusionment, it’s for the wrong reasons. 

The trough of disillusionment refers to the middle stage of the Gartner hype cycle, a concept that explains how people become acquainted, excited, and then disappointed with a technology before understanding it and, lastly, using it productively.

Lately the term has become common among those who claim that AI, despite its extraordinary promise, has yet to rescue burned-out physicians, control spiraling costs, or improve caregiving.

This perspective overlooks what it takes to make AI function in healthcare today. The real constraint is not the technology itself, but the time, deep clinical integration, and sustained commitment that’s necessary for AI to succeed in clinical settings.

The problem is that long-term effort is in short supply today not only in healthcare, but in the business models of most AI companies.

Healthcare is under extreme stress. That’s why healthcare leaders have high hopes for AI. But they frequently lack the institutional capacity to deploy the technology properly. At the same time, most AI companies are VC-backed and burning through cash. They can’t do it, either.

The problem is no longer theoretical: as healthcare leaders transition from buying clinician-facing AI to deploying it in real clinical environments, the gaps in implementation capacity are becoming impossible to ignore.

Deploying AI into real clinical workflows, not demos, not pilots that never graduate – is the hardest problem in healthcare right now. The technology exists. What’s missing is the willingness and ability to do the slow, unglamorous, clinician-by-clinician labor required to make it real.

Before an AI solution can be rolled out for a handful of caregivers, for example, scores of physicians must spend years training it to coordinate care plans for patients suffering from acute and chronic conditions, say, shortness of breath plus diabetes, and social determinants of health, like tobacco use.

These solutions must also take insurance, formularies, and other factors into consideration, or else physicians can’t adequately care for their patients. If they don’t include medical research and other tools, they also don’t really move the needle in terms of helping busy caregivers.

Health system leaders have shared a solution might require 30,000 validations that take days or longer to finalize before it’s ready for a pilot in a clinical setting. When a solution has the proper foundation, transitions from pilot to system-wide integration can take around three months.

Physicians in a single health system will then use the AI tens of thousands of times within a few months. That’s when the solutions learn more. The meticulous analytics and course corrections that follow are vital to their continued success.

Most health systems don’t have the expertise to manage this implementation and learning process alone, so they naturally lean on AI companies. But the capital structure behind most companies incentivizes shortcuts.

Companies race to demonstrate flashy capabilities, raise massive rounds, then pivot when the clinical deployment gets hard. This “demoware” looks impressive on stage but never reaches a real exam room.

Equity is a major culprit in this regard. When investors need returns on a timeline, companies optimize for the next funding round, not the next clinical validation. They chase the easier sale instead of solving thorny problems. Worse, when the market shifts or funding tightens, companies disappear, leaving health systems stranded.

Health systems have seen this cycle before and are understandably skeptical. Ironically, illustrating how AI is essential but hard to manage, many clinicians today are turning to so-called shadow AI, or personal AI tools, because they have lost faith that their institutions will adopt solutions to address their pressing needs.

Unleveraged companies, in contrast, avoid these burdens and devote themselves entirely to accumulating the clinical integration and trust necessary to bypass the trough of disillusionment. They can afford to put in the time. That’s the real moat in healthcare AI.

Deployment, not invention, is the bottleneck in AI in healthcare today. AI companies can build innovative, revolutionary solutions. The question is whether AI companies can undertake the hard, unglamorous work that will allow their solutions to change how medicine is practiced.