Table of Contents
Children’s Hospital of Philadelphia has trained more than 1,300 employees through its CHOP School of AI since launching the program in 2025, but its leaders say reaching the rest of its workforce will take more than curriculum alone.
How CHOP School of AI Built on Existing Data Literacy Work
The Philadelphia-based health system launched its School of AI in September 2025, building on a data literacy effort it has run since 2021 known internally as DnA University. That program has trained thousands of CHOP employees on data and analytics skills over the years, and CHOP used the same team and framework to stand up its AI-focused curriculum, according to Hojjat Salmasian, MD, PhD, vice president and chief data and analytics officer at CHOP, and Bimal Desai, MD, vice president and chief health informatics officer.
A Hackathon Kicked Off the Program
The first major undertaking under the School of AI was a hackathon that drew more than 100 attendees, paired with a set of recorded and live sessions on generative AI and how to use it responsibly.
CHOP School of AI’s Reach and Growth Targets
More than 1,300 CHOP employees have completed the health system’s more rigorous, role-specific AI training, Salmasian said, representing about 7% of CHOP’s roughly 20,000 employees. Basic, self-service introductory material has reached many thousands more, though CHOP does not count that toward its formal metrics. “Our target is to get to about 25% of the people working at CHOP, and we’re working sequentially towards that goal,” Salmasian told Becker’s.
A Structural, Not Resistance-Based, Challenge
Getting there is complicated by more than scheduling. Some departments show up more readily than others, Salmasian said, and the gap isn’t primarily about resistance. “It’s less so about resistance. It’s more structural,” he said. Research and operational roles tend to turn out more consistently than clinical roles, where any time spent on training competes directly with patient care and adds to a workforce already at risk of burnout.
How CHOP School of AI Reaches Clinicians Directly
To address that, CHOP has folded training into venues clinicians already attend, including grand rounds and department conferences, and built a network of internal “AI faculty,” clinicians and staff already comfortable with the tools who teach their peers directly.
A Peer-Led Teaching Model
“Basically, any time a clinical department or operational department requests, can you come to my team and teach us about AI, we’re more than happy to engage,” Desai, who is one of the AI faculty members himself, told Becker’s. “We have a lot of structured, predefined content that we can deliver, and we can tailor it to the specific needs of that group, whether they’re in finance or clinical care.”
Lowering the Barrier With Informal AI Stories
Some of that peer-led approach starts outside of work entirely. In his own division of general pediatrics, Desai said, monthly meetings now open with a colleague describing an informal use of AI, unrelated to their clinical duties, as a way to lower the barrier to trying it on the job. “We kick off with an AI story. Someone gives an example of how they’re using AI in their personal life, not necessarily in their work life,” Desai said. “Those examples I think kind of build this willingness to try AI, and if you’re willing to try it at home, maybe you’re more likely to try it at work.”
Measuring Success With the Kirkpatrick Model
To gauge whether the training is working, CHOP evaluates its programs using the Kirkpatrick model, a four-level framework that moves from learner satisfaction to measured outcomes, Salmasian said. The program’s early results are a 15.8% increase in understanding of responsible AI use and an 11.4% increase in confidence using AI tools, both self-reported gains that sit at the second level of that framework.
CHOP School of AI’s Impact on Usage and Burnout
Behavior change, the third level, is where CHOP has started tracking usage data directly. Some trainings have been followed by an eightfold increase in the number of employees using the AI or analytics tool covered in that session, Salmasian said.
Ambient Scribes Deliver the Clearest Measured Impact
The clearest example of the fourth level, measured impact, has come from CHOP’s ambient AI scribes. “The clinical scribes is probably our most well studied example of this, and I’ve personally been kind of blown away by the results of this tool,” Desai said. “We now have hard data from our hundreds of users at CHOP that this tool has resulted in a 54 percent reduction in self-rated burnout, which is invaluable to the institution and to those providers.” Patients and families have noticed too, he said, telling CHOP they appreciate that providers are no longer turning their backs to type during an encounter.
What Clinicians Are Saying About CHOP School of AI’s Tools
Some of the qualitative feedback has stuck with leadership as much as the numbers. Desai said providers have told him things like, “This tool has made me postpone my plans to retire,” or, “This is the best thing that’s happened in my clinical practice since I started medicine.”
Early Signals From Chart-Surfacing AI
CHOP has also tracked early signals from an AI tool that surfaces information from patient charts in outpatient and inpatient settings. One in four providers using it report learning something new about their patient, Desai said. “We don’t have the hard data yet to show that this is going to drive quality or safety outcomes, but you have to imagine that if you do this at scale, people are discovering things about their patients that are actually critical for care delivery,” he said.
What CHOP School of AI Leaders Recommend for Other Health Systems
Asked what advice they’d offer other health system leaders trying to build similar AI education workforce programs, Salmasian said literacy is only the first step. “While training can be the reason that you jumpstart somebody’s use of a tool in analytics or AI or other spaces, it’s not going to get you necessarily to continuous adoption, and that requires a different level of continuous support at the elbow, support champions, and a mixture of other tactics and strategies that you can use to make sure that people are actually continuing to harness the value of the tool or capability that has become available to them,” he said.
Bringing Implementation Science to Future Rollouts
CHOP is now working with its own Clinical Futures Research Center to bring implementation science methods to how it rolls out and supports new AI tools, Salmasian said, calling adoption a persistent challenge even for popular tools like the AI scribes. Desai added that meeting employees at their own comfort level has been central to the program’s design. “You end up meeting users where they are,” he said. “You can start at a very basic level, let’s talk about what is a prompt and how do you construct a good one, the prompt engineering 101, all the way up to really sophisticated examples where they’re creating agents that can do tasks on their behalf and things like that, and everything in between.”
What This Means for Health System AI Training Programs
CHOP’s experience with its School of AI suggests that sustainable AI literacy in healthcare requires far more than one-time training sessions, demanding ongoing peer support, structural accommodations for clinical schedules, and rigorous measurement frameworks like Kirkpatrick to track genuine behavior change rather than just satisfaction scores. As CHOP continues working toward its 25% workforce training goal and partners with its Clinical Futures Research Center to apply implementation science methods, its approach may offer a useful model for other health systems trying to move beyond curriculum alone toward lasting, measurable AI adoption across both clinical and operational roles.
For more healthcare industry updates, insights and news, visit DistilINFO. Click here to subscribe to stay informed.
