Hospital AI Pilots Work. Scaling Doesn't.

A hospital picks one floor. One champion, usually a physician who actually wants the thing to work. One narrow workflow, say, an AI tool for a single documentation task. Six months in, the numbers look good. Time saved. Adoption inside that unit is high. Leadership is pleased.

Then someone tries to roll it out past that one floor. And it stalls.

This is not a rare story. A new survey of healthcare leaders, run by Carta Healthcare and covered by Healthcare IT News on August 7, found that 71% of organizations that have already proven measurable value from an AI tool are not expanding it at pace. The pilot works. The scale-up does not.

It was never about trust

The easy explanation is that doctors do not trust the machine. That story is getting old, and the data does not back it up. In the same survey, only 26% of respondents named clinician trust as a barrier to adoption. Compare that to 44% who named one specific thing: difficulty integrating with the electronic health record.

Brent Dover, Carta Healthcare's CEO, put it plainly to Healthcare IT News: "The pilots proved that it does. What stalls is everything that comes after the proof."

He is right about why. A pilot works because it is small. One physician champion. One well-defined workflow. Concentrated attention from a team that wants it to succeed. Roll that same tool out across a department, and every one of those conditions disappears. What is left is a tool that lives outside the chart, asking a doctor to open another window, log into another system, copy something back into the record they are actually responsible for.

That is not a trust problem. That is a workflow problem, and it shows up as friction, not as skepticism.

The tell in the data

The survey's respondents were asked what would most help AI adoption move faster. The top answer, at 48%, was a solution that lives inside the EHR itself. Not more evidence. Not more peer case studies, though those helped too, at 36%. A tool that is already where the work happens.

Dover said it this way: "When a solution lives inside the EHR and proves itself against measurable outcomes, adoption follows."

That is the whole argument, really. Doctors do not reject AI because they distrust it. They reject a second app, a second tab, a second login, bolted onto a shift that already runs on borrowed minutes. A pilot can absorb that tax because it has a champion covering the cost. A hospital-wide rollout cannot, because nobody is covering it for three thousand doctors at once.

Why this matters past one survey

None of this is really about Carta Healthcare's product, or any single vendor's product. It is about where the tool sits relative to the doctor's actual shift. A point solution, however good its underlying model, is still one more thing to open. A workspace that already holds the patient list, the notes, and the handover is a different kind of object. It does not ask the doctor to go somewhere else to get value from it.

That is the bet behind Medows, and it is why this survey is worth reading past the headline. The industry has spent two years arguing about whether AI is accurate enough for the ward. The more useful question, this data suggests, is whether it is close enough. Close enough to the chart, to the round, to the ten minutes a resident has between patients. Accuracy without proximity gets you a good pilot. It does not get you a hospital that has actually changed how it works.

Sources

• Hospitals must prove they can make AI operational at scale, Healthcare IT News, August 7, 2026
• 92% of Healthcare Leaders Won't Trust AI at Scale Without Deep Clinical Expertise, Carta Healthcare survey via PR Newswire, July 21, 2026