AI scribe evaluation: trial design and metrics

In Episode 27, the AI or Die crew is back.

The team is joined by special guest Jenny Owens, Senior AI Program Administrator at Cleveland Clinic and Chair of the HIMSS Northern Ohio AI Center of Excellence Committee. Jenny brings a rare, inside look at how a major health system evaluates, validates, and governs AI — from the human realities of working with LLMs to running a full AI scribe evaluation that began with 126 vendors. Her perspective offers one of the most practical views of modern AI governance you’ll find.

Show Notes

0:00 – Welcome Back Rehgan returns from maternity leave; team catches up on Halloween abroad, newborn sleep, LA travel, and what changed (and didn’t) in AI during two months away.

4:30 – Industry News – OpenAI’s IPO ambitions – NVIDIA & Palantir partnership momentum – California’s proposed ban on AI posing as licensed health pros – Ohio’s proposal to outlaw AI “marriages” – Public anxiety around therapy bots, critical thinking, and “AI psychosis”

13:20 – AI in Medicine Parenting + AI tools; physician reactions to AI-influenced patient questions; why clinicians (not laypeople) are best positioned to validate AI outputs.

18:45 – Enterprise AI Shifts AI flattening org charts, reducing communication bottlenecks, driving efficiency trends (including Amazon layoffs), and reshaping cross-team workflows.

22:30 – Guest Joins: Jenny Owens Jenny’s roles at Cleveland Clinic and HIMSS; host of the Health Data Ethics podcast.

23:00 – Governance + Enablement Her dual remit across governance and AI literacy reduces friction and creates holistic oversight. She describes governance mindsets (“Sam the Eagle”: risk & rules) vs. innovation mindsets (“Doc Brown”: speed & experimentation) — and how her job is to align both.

28:30 – Cleveland Clinic’s Governance Structure A multidisciplinary council including clinicians, cybersecurity, legal/compliance, data science, and bioethics. Reviews cover model performance, contextual risk, intended use, human interpretation, workflow impact, and ethics. Research AI is evaluated separately but reviewed early if headed toward clinical use.

34:00 – Culture & Values Governance mirrors Cleveland Clinic’s culture of curiosity, teaching, and patient-first thinking. Teams zoom out to philosophical questions: What does accuracy mean for LLMs? What hallucination tolerance is acceptable? How should non-determinism be handled? Decisions stay contextual and case-specific.

41:00 – Governance’s Core Challenge AI governance must evaluate both the model and the human using it. Few orgs have mature systems to monitor model performance alongside human actions. Governance is shifting from model-centric to workflow-centric.

44:20 – “AI Candyland” Jenny’s vision: a guided, gamified pathway for employees exploring AI ideas — with defined starting points, bumpers, fast tracks, and off-ramps. Designed to empower clinicians while keeping systems safe.

47:30 – The 126-Vendor AI Scribe Evaluation Cleveland Clinic began with 126 vendors, narrowed to 5, and ran a full evaluation planned for 6 months that ultimately took 12. Metrics included statistical performance, workflow fit, burnout impact, billing/coding reliability, draft quality, edit burden, and discrepancies between perceived vs. actual time saved.

54:00 – Why AI Projects Fail Common patterns: vague requirements, no baselines, poor change management, overhyped promises, lack of co-design, and leadership mandates without clarity.

56:00 – Wrap-Up Connect with Jenny on LinkedIn and via the Health Data Ethics podcast. The team signs off before a short holiday break.