
253: Dayton Children’s CIO JD Whitlock: What Would Make a Health System Switch AI Scribe Vendors, Why Most AI Is Theater, and the Pediatric AI Gap Nobody Is Funding
- Written by
- SeamlessMD
- Published on
Subscribe on: RSS | SPOTIFY | APPLE PODCAST | GOOGLE | BREAKER | ANCHOR
On this episode of The Digital Patient, Dr. Joshua Liu, Co-founder & CEO of SeamlessMD, and colleague, Alan Sardana, chat with J.D. Whitlock, Chief Information Officer at Dayton Children's Hospital, about "What Would Make a Health System Switch AI Scribe Vendors, Why Most AI Is Theater, the Pediatric AI Gap Nobody Is Funding, and more..."
Audio:
Guest(s):
- J.D. Whitlock, Chief Information Officer at Dayton Children's Hospital
- Joshua Liu, MD (@joshuapliu), Co-founder & CEO at SeamlessMD
Episode 253 - Show Notes:
[00:06:35] How Dayton Children's chose Abridge, and what could make it switch to Epic's ambient tool — in summer and fall 2024, JD wanted an ambient tool integrated with Haiku, which early adopters said was key to mass provider adoption on Epic. That narrowed the field to Abridge and Nuance/Microsoft, and with Microsoft not yet having cut its price from the human-in-the-loop DAX, "the decision somewhat made itself." As Abridge's standard three-year contract comes up, the team will have to weigh Epic's own "chart with Art" option.
[00:08:18] Why the usual "Epic version two is 90% as good" math may not hold for ambient — health systems often wait for Epic's second or third version of a best-of-breed product, then switch to save money and gain integration. Because ambient is "such a huge deal for provider satisfaction," and Abridge is so well funded and moving so fast, providers may not accept a switch until there's real feature parity. On the other side, the full agentic visit experience Epic demoed at UGM relies on Epic's ambient tool. JD notes that the early adopters moving to Epic's ambient were all coming from DAX: "Nobody was coming from Abridge or Nabla."
[00:10:39] How a 30-provider volunteer group keeps up with a flood of new AI features — Dayton Children's has 300 Abridge seats and is on Abridge's early adopter program, pushing new features to 30 volunteer providers as soon as they arrive. JD calls the pace "a good problem to have," though it can be challenging to absorb "five new features this week."
[00:11:55] Why AI governance asks half the questions of trusted platform vendors — anything generative, and predictive AI that clinical decisions will rely on, such as sepsis prediction, needs an additional governance layer, which JD says is nothing new on the predictive side. Epic customers quickly share word when a new tool isn't ready, and features coming from trusted platform vendors like Epic, and increasingly Abridge, carry a trust factor: "You only have to ask yourself half the questions" compared with a net-new third-party bolt-on.
[00:14:01] What a CIO needs from AI vendor pricing models — with 1% margins, health systems need predictable budgets and can't add significant costs mid-fiscal year, which is why software was long priced on hospital beds. JD says it's fair that vendors charge for orders of magnitude more compute, much like SMS platforms charge per million texts at a cost so low no one hesitates to run a campaign. Epic's AI pricing has gone from variable to fixed and back to variable, but for Dayton Children's it's "not gonna break the bank," while Workday's annual "flex credits" cover both generative AI compute and API usage.
[00:17:10] Why giving 1,000 people agent-building tools is probably not ROI — unlike software companies that can justify going over budget on tokens, Dayton Children's isn't developing software. JD sees good agentic potential in data and analytics and has given data analysts Microsoft E7 licensing with agentic orchestration features, but cautions that letting 1,000 people build agents to go through their inboxes "is probably not ROI." Token consumption budgets aren't yet a conversation at Dayton Children's, though academic medical centers may face different pressures.
[00:21:57] What "AI theater" looks like, and what Dayton Children's is choosing not to chase — JD isn't chasing agentic automation unless it passes a normal ROI calculation for the new labor involved. AI theater, he says, is the Becker's article where an organization announces how smart it is for using AI, "and they just turned on that Epic feature that everybody else has turned on." Some health systems also face board pressure to replace whole departments with AI. On why healthcare celebrates implementations at all, JD adds: "You can kill patients if you do it wrong."
[00:24:54] How a smaller health system weighs voice bots: Epic Emmie, a healthcare vendor, or Silicon Valley — options range from Epic's Emmie, to healthcare-only vendors like Hyro, to big non-healthcare players like Decagon or Sierra. Beyond the vendor choice, JD frames a cultural question: "Are we ready to have bots talking to our patients?" Health systems must decide between tight Epic integration that flows into pre-visit planning, the most natural human-sounding voice, or simply using AI to get callers to the right human faster than a phone tree.
[00:26:41] Why Dayton Children's passed on AI fax-referral automation and chose to wait — rural and smaller practices still fax referrals, and vendors now combine AI and OCR to ingest them. But those vendors cost "many multiples of the labor cost" of the couple of people typing referrals in. Instead, Dayton Children's is implementing Epic Gallery, Epic's document management solution, expecting it to include AI-enabled OCR by go-live: "the smart thing to do is just wait a year."
[00:28:32] Why waiting on AI carries real competitive risk in patient experience — doctors often believe patients are loyal, but data shows many would switch for easier scheduling and a more convenient journey. If one of two competing health systems transforms its patient experience with this technology, "that could be a meaningful competitive difference." Even for considered decisions like surgery, a website that matches patients to a surgeon by insurance, language, complication rate, satisfaction scores, and a short introductory video can set a system apart.
[00:31:14] How answer engine optimization and smarter "find a doc" tools help low-literacy patients — Dayton Children's marketing team is "all over" answer engine optimization. JD has used Kyruus since his previous role at Bon Secours Mercy Health to match patients who misspell a diagnosis or procedure with the right physician, such as a hand surgeon rather than any orthopedic surgeon. Asking someone who didn't finish high school to know what a nephrologist does "is actually not a good find a doc experience."
[00:33:31] Why "children aren't small adults" applies to AI products too — the largest category of FDA-approved AI is medical imaging, yet there's very little for pediatrics. Dayton Children's is waiting on a promising bone age tool that isn't yet FDA certified, while adult systems on a modern PACS use many AI imaging tools. JD's advice to vendors: "demo it with a pediatric disease instead of heart failure."
[00:34:47] How video visits added a new dimension to adolescent confidentiality — states set different age limits, and parents' email and phone numbers often sit in their child's MyChart. A parent of a 15-year-old who needs to talk privately with a pediatrician knows the appointment is happening and pays for it, but shouldn't see what was said. Before COVID, doctors sent the parent into the hallway; with video visits, the doctor sees only the teen holding the phone and has to wonder whether mom and dad are listening.
[00:36:21] Why pediatrics doesn't need its own ambient tool, but does need more investment — in demos with staged interruptions, ambient AI proved good at ignoring non-medical conversation, like small talk about a football game or a noisy sibling. JD says children are a quarter of the population "but it's not a quarter of the money that we spend on healthcare," which explains the lack of investment. Pediatric opportunities do exist, such as solutions for safely sending infants home from full NICUs with monitoring, and startups coming through KidsX, the accelerator run out of Children's Hospital Los Angeles.
[00:39:50] How "imitate to innovate" helps a smaller system tell what's truly proven — most Epic AI tools work fine in pediatrics, but a couple needed a second or third version first. JD watches larger pediatric academic medical centers: "if it works at CHOP, it's probably gonna work at Dayton Children's," and if CHOP says a tool isn't ready, they wait. Scale and ROI can still differ, but normal diligence catches that.
[00:40:57] Why open sharing among non-competing health systems is a hallmark of the Epic ecosystem — presenting at UGM or XGM is "a feather in your cap," but it comes with the expectation that you'll share, often with a line of people waiting afterward. Most health systems compete directly with only a handful of others, so the remaining 500–600 share freely. CIOs routinely ask peers to have their teams spend 30 minutes sharing lessons learned, "and then, of course, I will, of course, return the favor."
[00:43:43] What KLAS Pinnacle Awards and 97th-percentile scores do and don't mean — Dayton Children's earned its KLAS Arch Collaborative results by surveying over several years and fixing what doctors and nurses were unhappy with. JD likes that the Pinnacle threshold is fixed rather than rising each year, but cautions that a 97th percentile score "does not mean 97% of our providers are blissfully happy with their Epic experience." The next survey will be the first since implementing Abridge.
[00:45:22] How 20 clinical governance committees avoid becoming 20 veto points — JD argues the average Epic customer needs roughly that many committees to bring the right clinical and operational leaders to decisions across modules. Applying the theory of constraints, "the most useful thing that I learned in business school," means watching which certified Epic teams, often interfaces and access, are overloaded. Medium-size optimizations that fit existing staff and licensing go to lower-level committees, while new budget requests go up to a senior committee with budget authority.
Fast 5 Lightning Round:
1. What is your favorite book or book you've gifted the most?
Falling Upward by Richard Rohr
2. If you could instantly master any skill, what would it be?
"Become a virtuoso on my red Gretsch guitar"
3. Would you rather have super strength, super speed, or the ability to read people's minds?
"Absolutely read people's minds. That'd be the most fun."
4. What is something in healthcare you believe that others might find insane?
"We have to get a handle on end of life care"
5. What is the last movie or TV show you saw, and what did you think of it?
"Mayor of Kingstown. We're horrified by each episode, and then we can't wait to watch the next one."
The Digital Patient has been recognized as Feedspot's #1 Patient Engagement Podcast of 2025. Thank you to our listeners for making this happen!