Small changes, huge stakes

In eighteen forty-seven, Ignaz Semmelweis asked Vienna maternity clinic doctors to wash hands in chlorine solution before examining patients. Deaths fell from about eighteen percent to about two percent within two months.

“In healthcare, small changes in practice can have disproportionately large effects, because the healthcare system affects everyone and the stakes are literally life and death.”

Why handwashing matters now

Healthcare changes affect everyone and stakes are life and death. AI could revolutionize healthcare, but institutions are failing to adjust despite widespread interest from patients and providers.

Patients are already using AI

Roughly a third of Americans have used AI for health information. One in four did so in the past month through chatbots like ChatGPT. Younger people and uninsured patients use these tools most.

The risk of unmet demand

People want more healthcare access and are finding creative ways to meet their needs through AI chatbots. When a chatbot is wrong, consequences fall on people least able to absorb them.

“This is a form of unmet demand, clearly people want more access to healthcare.”

Doctors adopting AI independently

Doctors are adopting AI mostly on their own. OpenEvidence, a free clinical reference, skyrocketed from about three million physician consultations a month to about eighteen million a month in December twenty twenty-five.

System-wide adoption is needed

Individual doctor experimentation is normal and expected. But significant benefits for society require effective, system-wide adoption of AI in healthcare institutions.

AI adoption is failing everywhere

Healthcare is not alone. Across industries, eighty-eight percent of AI pilots never make it to production. Only about one in ten companies report any productivity gains from AI over the last three years.

The gap between hype and reality

This gap is huge and at odds with spending and media attention surrounding AI. One main impediment is something leaders can address today: they have the wrong mental model for AI.

Proven track record

Over nine years, the authors' firm worked on more than two hundred thirty AI projects across thirty industries. Ninety-six percent of models hired to take to production shipped, including in healthcare.

The wrong mental model

Leaders treat AI like traditional software, when it is not. This wrong mental model is a critical blocker to progress, just as Semmelweis's colleagues rejected handwashing for decades despite evidence.

“They are treating it like traditional software, when it is not.”

How AI differs from software

Traditional software is written; AI is trained and grown. Software is deterministic: if this, then that. AI is probabilistic and dependent on its data. Software is built linearly; AI is built in loops.

The production problem

Most AI work happens before anything resembling a product exists. Managing AI like software means specifying the answer before checking whether the data can actually produce such an answer.

“Managing AI like software means specifying the answer before you've checked whether the data can actually produce such an answer.”

Healthcare must change its approach

Healthcare is complex, high-stakes, and its data is fragmented across institutions reluctant and often legally unable to share. Leaders must adopt a new mental model for AI to unlock its potential.