While healthcare AI proliferated in 2025, health systems decided to dial back on the technology for some uses.
The strategy foreshadows the approach that healthcare organizations will take in the new year and beyond, as AI applications will have to start showing results — or risk getting shelved.
“We retired a professional billing tool that was supposed to make coding easier for providers, but the implementation was far more complex than what we were told and we were able to terminate the contract prior to go-live,” said Will Landry, senior vice president and CIO of Baton Rouge, La.-based FMOL Health.
Coral Gables-based Baptist Health South Florida recently dropped an AI-powered virtual care navigator designed to guide patients to the appropriate care level or venue, said Chief Digital and Information Officer Sha Edathumparampil. Despite the platform’s potential, fewer than 100 patients engaged with it monthly. In 90% of cases, it suggested urgent or emergency care when more suitable options were available.
“This experience highlights key lessons about the state of healthcare AI in 2025. While the field is maturing, it demands robust human oversight, iterative testing, and alignment with real-world clinical and operational workflows to deliver true value without overburdening systems,” Mr. Edathumparampil said.
Duarte, Calif.-based City of Hope stopped using commercial AI for clinical documentation improvement because of information security and data privacy concerns, instead building a successful solution internally that the organization may bring to market, said Simon Nazarian, executive vice president and chief digital and technology officer.
“What it taught me is to continue the level of scrutiny and make sure we provide ample time and investment for AI and data governance capabilities,” he said. “There is a lot of automation and predictive analytics out there that are being presented as AI (generative or agentic). We need to keep a sharp eye and make sure our investments are curated properly to focus on real impact rather than the ‘shiny object.’”
Oskaloosa, Iowa-based Mahaska Health reversed its entire AI game plan in 2025, said CIO Bob Berbeco. “We stepped away from running isolated, bolt-on AI pilots that were not proven by being embedded in real clinical workflows or were likely to be duplicated by Epic’s AI roadmap,” he said.
The one-hospital system also established a Data Science and AI Center of Excellence to formalize standards and its intake process for new tools.
“Governance, security, measurement discipline and strong data foundations are what separate short-lived pilots from sustainable scale, particularly as core EMR platforms rapidly expand their native AI capabilities and make potentially redundant point solutions easier to avoid,” Mr. Berbeco said.
While Charleston, S.C.-based MUSC Health didn’t discontinue any AI products in 2025, it was planning to use a smaller vendor’s denials tool before switching gears at the last moment and opting for Epic’s version, which had come out quicker, said Chief Digital Transformation Officer Crystal Broj.
At Walnut Creek, Calif.-based John Muir Health, some AI pilots took longer than anticipated, said Priti Patel, MD, vice president and chief medical information officer. The reasons: unexpected results, poor adoption, outcomes that fell short of projections, cost considerations.
“However, we have chosen not to retire these projects because the vendors seem to be continuously iterating and improving based on our feedback,” Dr. Patel said. “We’ve also found that successful adoption of AI takes time and effort. Although these projects haven’t been as successful as we had hoped, they have provided us with valuable insights into implementing AI.”
Columbus, Ohio-based Nationwide Children’s Hospital didn’t decommission any AI platforms but rebooted its service desk chatbot, said CIO Denise Zabawski.
“We had not put the correct operational processes in place and the opt-out portion failed, providing poor customer service,” she said. “We’ve also been slow to deploy Copilot and ChatGPT connectors until we complete internal security work on our file shares.”
Mishawaka, Ind.-based Franciscan Health’s use of AI was limited in 2025 compared to similarly sized peers, so the 12-hospital system is still testing out — and not sunsetting — use cases, said Chuck Christian, vice president of technology and chief technology officer.
“We have been working on our governance structure and processes over the last many months to ensure that we’re taking a measured approach to the implementation of AI within our enterprise,” he said. “I’m betting if you ask the same question this time next year, I might have a different answer for you.”
