of health systems are running or planning AI pilots.
of piloting systems are testing anything that informs a clinical decision.
ambient documentation pilots is described as finished.
AI buyers have no way to monitor how their AI performs.
At one large system, about 1 in 5 pilots reached full scale.
Five reasons pilots stall
No one owns the decision. 13 of 30 organizations raised unclear AI ownership without being asked.
Review can't keep up. One use-case review took 6 to 10 weeks. Reviewing one EHR vendor's catalog of 64 tools would take about four years.
Nobody budgets for scale. Vendors pay for the pilot. The health system pays for the rollout, and only 1 of 36 buyers has a budget for managing AI.
No one can see how the AI performs. 19 of 36 buyers can't monitor AI performance, and 10 of 30 organizations say vendors withhold performance data.
Every graduation is a custom integration. 20 of 29 systems run two or more data warehouses.
Part 3 lists 10 practices health-system leaders described to us. Two examples.
- Agree on success before the pilot starts, and confirm the data to measure it will be there.
- Tell the sponsor the full cost at scale before the pilot begins.
"My job description doesn't have AI governance in it. There's no one in our institution that really has that kind of title or ownership."
"Everybody says you have to monitor AI programs, but nobody really does know how."
Asked how accurate its ambient listening was, a vendor answered: "Yeah, no, go pound sand."
Grid surveyed IT and security leaders at 29 U.S. Epic health systems (22 with 1,000+ beds, 7 under 1,000) and held recorded conversations with 36 health-system AI buyers. Interview figures are each leader's account of their own organization and are labeled that way in the report. Leaders are identified by role and organization type only. Published November 2026.
- Key findings and methodology
- Part 1. Identifying the gap
- Part 2. Why the gap exists
- Part 3. Closing the gap: 10 practices
- Conclusion
Grid Advisory helps health systems set up AI ownership, review and pilot criteria. See Professional Advisory
