Partner Perspective: Drawing on Global Experience, Heidi Health’s Kelly Offers Advice on Optimizing Clinician AI Adoption

Health system IT executives tasked with diffusing AI into their organizations are facing a serious challenge: how to ensure they get behind the right tools, those that will actually make a difference in clinicians’ lives by helping them become more productive at work and giving them more time at home.

In a wide-ranging discussion, Dr. Thomas Kelly, Co-Founder & CEO, Heidi Health, outlined a playbook for moving beyond proofs of concept to broad, durable adoption—one that starts with super-users, measures activation rather than licenses, embraces workflows inside and outside the EHR, and anticipates emerging regulatory obligations tied to clinical reasoning. Kelly argues that systems that operationalize adoption mechanics early will see the compound benefits in throughput, documentation quality, and clinician sentiment. It’s the approach his company uses and finds much success with around the world.

Building Around Super-Users and Specific Workflows

The center of gravity, Kelly said, should be a cross-department council of clinically credible champions who help map where AI actually fits day today. The area where AI can help is deliberately broader than EHR notes and covers multidisciplinary team meetings, operative documentation, form summaries, results synthesis, and pre-visit preparation—often the cognitive work that precedes the formal chart. In his experience, adoption rises when each specialty sees its own use cases reflected in the initial configuration, rather than receiving a generic scribe. “One of the most important things that we find is bringing together what is almost like a council or committee of super users across organizations.”



From there, the team inventories workflow friction service-line by service-line and builds templates, examples, and short demos clinicians can reference on day one. Kelly favors a train-the-trainer cascade—champions teach ten, ten teach a hundred—supported by self-service materials and department-specific content. That structure, he noted, reduces hand-holding and normalizes use across varied clinical settings where the value of AI may differ.

Measure Activation, Not Pilots

Kelly notes that in other parts of the world, where Heidi is an established and popular tool, success as a vendor depends on customer use and adoption. For example, in order for pilots in other parts of the world to expand, a majority of eligible clinicians must be using the tool. “We have to get at least 80% of the clinicians using it, at least once a week, ideally every day.”

He cautioned that buying a block of seats or running a small pilot can mask low real-world engagement; the measure for CIOs is how quickly usage concentrates among practicing clinicians.

Activation, in this framing, is not just a dashboard number; it is a gate to expansion. Kelly argued that systems that insist on explicit exit criteria for alpha and beta phases—weekly-active thresholds, specialty coverage, and satisfaction targets—avoid the risk of a “switch-on-for-everyone” launch before workflows are ready. He added that specialties vary widely in terms of where AI helps so activation must be read locally and managed accordingly.

Design for Use Inside and Outside the EHR

According to Kelly, adoption falters when tools are welded to the record and optimized for primary care alone. Many specialties do their highest-value thinking before a note, and much of that work—triage synthesis, evidence review, and handovers—doesn’t require immediate insertion into the chart. Enabling clinicians to use AI both within the record and as a companion alongside the record expands the surface area for value and increases the frequency of daily use. That flexibility, he said, is why cross-specialty activation can exceed 80% when workflows are mapped first and integration follows the use case rather than dictates it.

Kelly advised CIOs to encourage departments to start where value ...