Health systems have lately been managing token-based AI costs across their broader vendor stack largely through access tiering and governance.
What emerged at this year’s Epic Users Group Meeting is a narrower problem: The vendor’s own AI suite, now woven directly into the EHR nearly every clinician and biller touches daily, runs on a hybrid pricing model that CIOs say is uniquely hard to isolate, forecast or govern the way they can with standalone AI tools.
Several CIOs used the same word for it: tokenomics.
“The other piece that I think is interesting right now is all the consumption-based AI costs with Epic and others, and how we in healthcare figure out how to plan for that,” Amy Trainor, BSN, RN, vice president and CIO of New Orleans-based Ochsner Health, told Becker’s at Epic’s headquarters in Verona, Wis. “It’s so variable that we’ve got to really sit down [and] make sure we have accountable people to understand: Are we getting the ROI and benefit from each of these pieces of AI? … That’s a whole new function of accountability and job function that just doesn’t exist today.”
Ms. Trainor said Epic’s own AI features are billed inconsistently — some flat, some consumption-based — and that Epic has only recently begun building the kind of per-feature cost dashboard that would let a health system see what it’s actually paying for. “Each feature, hundreds of features, will show what that looks like. It could be $10. It could be $10,000, depending on what the usage and the price model looks like,” she said. “Planning a budget a year in advance is a little more difficult than it’s been in the past.”
There’s no cushion if a health system misjudges demand, either, she said. “I don’t imagine there’s … a refund policy. Like, ‘Hey, you guys, sorry you missed it.’ … I just don’t envision that.”
That uncertainty is reshaping headcount, not just spreadsheets. Ms. Trainor’s organization, like others, is now building out roles that didn’t exist a year ago — “more QA, more engineering, financial FinOps for each team that has a consumption piece,” she said, describing the job as making sure “we’re not blowing our budget out of the water” on any given AI feature.
Aaron Miri, DHA, chief digital and information officer of Jacksonville, Fla.-based Baptist Health, said the industry still lacks a clear way to weigh AI’s return against what it costs to run. “Each of those … tokens have a true cost,” he said. “If we go hog wild with [Agent] Factory and others, which we intend to, there’s a cost of doing business. You may have savings on the operational side — great, thumbs up. Who’s paying the piper on the back end, paying for the tokens? We’ve got to work through that … sausage-making.”
Dr. Miri said that calculus extends to infrastructure decisions Epic doesn’t fully control: whether to run AI models in Epic’s cloud or on health systems’ own servers to avoid per-token fees. “We’ve also got to have some on-premise [capacity] to be able to run some of those models locally, so we do not incur those costs,” he said. “Now there’s GPU costs on the other side, and we’ve got to consider [that]. It’s a calculus … we just haven’t gotten there yet as an industry.”
Even figuring out whose budget absorbs the bill is unresolved. Asked by a fellow CIO at UGM where the money for AI comes from, Dr. Miri said Baptist Health has landed on a workaround rather than a clean answer: “I haven’t had full support to [move] money out of operations to give it to IT. We’re doing a shared [approach] — any money coming out of operations goes into a shared pot. A portion of that’s reallocated back to IT, but it’s not clean yet.”
Sha Edathumparampil, chief digital and information officer of Coral Gables, Fla.-based Baptist Health South Florida, pointed to the structure of Epic’s pricing itself as the source of the confusion. “The pricing model is super interesting. If you look at some of the competing products, [ambient listening tools are] one license fee,” he said. “But Epic, because of rising token costs and the economics around it, [has] a fixed fee and then a variable … [charge], which depends on which frontier model you use. It’s only increasing — some increase faster than others, so that’s a concern.”
He said Epic’s move toward supporting open-weight models, discussed during this week’s sessions, could ease some of that pressure — but he wasn’t yet sure by how much. “That should help address some of those cost concerns from a token-cost perspective,” he said.
The uncertainty isn’t limited to back-office tools. Maggie Helms, senior vice president and chief data, AI and digital officer of Bloomington, Minn.-based HealthPartners, said the economics of Epic’s patient-facing AI agent, Emmy, are still an open question as the tool’s readiness expands. “It’s great to be available after hours … for as long as the patient likes,” she said. “But then how do we monitor the costs [and] direct them better? That’s something I was hoping to hear a little more about.”
At Somerville, Mass.-based Mass General Brigham, Eric Podradchik, vice president of digital clinical systems, said the cost conversation extends beyond individual AI features to the infrastructure Epic now expects health systems to run underneath them. Epic’s cloud data platform, Cogito Cloud, has effectively become mandatory, he said, “much like Cosmos was before. … You’re a customer, and you’re being forced to implement something to maintain your current platform. At the same time, it’s providing the capabilities that you want, so it’ll be interesting to see how they develop it and how that translates to impact from capabilities as well as system sustainability and cost management.”
Epic executives acknowledged the visibility gap during the conference. The company said this week it’s building individualized cost-tracking and forecasting tools similar to what Microsoft already offers its own enterprise customers — a sign, Ms. Trainor said, that Epic recognizes the problem, even if the tools aren’t fully built yet. “Epic doesn’t have quite [that] all the way down to the user,” she said. “Right now it’s by function … and by time.”
For now, several CIOs said, the honest answer to how much Epic’s AI suite will cost their organization next year is that they don’t know — and won’t until the tools to measure it catch up with the tools generating the bill.
At the Becker's 11th Annual IT + Revenue Cycle Conference: The Future of AI & Digital Health, taking place September 14–17 in Chicago, healthcare executives and digital leaders from across the country will come together to explore how AI, interoperability, cybersecurity, and revenue cycle innovation are transforming care delivery, strengthening financial performance, and driving the next era of digital health. Apply for complimentary registration now.
