Five takeaways on AI governance, pilots, and virtual care from CHIME’s Renton Innovation Summit

Key takeaways

Leaders at CHIME’s Renton Innovation Summit argued that health systems innovate by cutting applications and ventures, fixing processes before buying technology, and ending pilots on schedule. Virtual care is now a workforce strategy as much as an access strategy, and AI governance works best when it finds a safe path to yes while keeping human reviewers from going on autopilot.

At the most recent CHIME Innovation Summit, held Oct. 14 to 16 at Providence Health & Services headquarters in Renton, Washington, host Chero Goswami opened by telling health IT leaders to stop counting the seconds their tools save. Goswami, Providence’s chief information and digital officer, said a tool that cuts 27 seconds from a clinical note has returned, in his words, a fistful of sand from a beach. He asked instead whether the physician who was charting at 9 p.m. now eats dinner with family four nights a week.

He posed the same question to every session at the summit. How is this product or process giving time back to the person on the other side of the technology? “We’ve been disrupting everybody else for 30 years and calling it innovation,” he said. “Let’s disrupt ourselves.”

Goswami’s question hung over the rest of the summit, the last of CHIME’s 2026 series, which drew Providence’s operating, finance, and clinical informatics leaders along with counterparts from Stanford Health Care, MultiCare, UCHealth, Sky Lakes Medical Center, TriState Health, TrueCare, and others. The leaders agreed that obstacles to innovation were mostly internal, but debated how far to trust AI with the repair.

Five takeaways from the Renton Innovation Summit

Leaders counted retiring applications and spinning down ventures as innovation

The opportunities in fro nt of health systems have not changed in 27 years, Leslie Flake, Providence’s CFO for system and operations finance, told the room. Five financial levers matter: revenue cycle, productivity, length of stay, medical group access, and supply expense. Every organization she has joined ran four times the applications it needed, which she blamed on governance rather than money. “Finance is secondary,” she said.

“My house is on fire, but I’m out in the backyard building a pool,” said Darryl Elmouchi, Providence’s COO, describing systems that innovate before fixing the basics. His own system’s proudest recent work was the unglamorous kind, he said. Providence cut its portfolio companies from about 60 to fewer than 30. Retiring software is harder than buying it, added Scott Waters, MultiCare’s interim CIO. Every application MultiCare drops has to be pried loose, so it now reviews new requests jointly with its CFOs.

Online scheduling keeps failing because the process behind it never changed

The biggest failure of Elmouchi’s career, by his own account, was a scheduling rollout at a prior health system. The cardiovascular service line volunteered to go first, errors piled up, and he pulled the system out after nearly a year. “We tried to digitize an inherently analog process without changing the process,” he said. Success later meant collapsing 300 visit types to 12 and getting physicians to accept imperfect matches.

Leah Rosengaus, executive director of digital health at Stanford Health Care, said her team spends most of its time getting operations to agree on a workflow rather than configuring tools, because surgeons’ scheduling rules still live in notebooks. Moving work from one group to another is not innovation, said Deb Carpenter, TriState Health’s CIO. “You’re just shifting the workload.”

Pilots need an end date and a scaling path before they start

Elmouchi described how Providence now runs a pilot. It gets a window, often 60 days, with success defined up front, and then it either scales or stops. Providence also runs several picture archiving systems, so it will not pilot radiology AI until it has one.

Sepsis care supplied the day’s clearest example of a pilot that scaled. Brittany Cyriacks, clinical informatics program director at UCHealth, said the system routed sepsis alerts away from bedside nurses to a command center, where a virtual ICU nurse reviews the chart and drives the treatment bundle. When UCHealth acquired hospitals with poor sepsis outcomes, it added them to the same logic the next day, and she said they moved from the bottom decile to the top.

Health systems now lean on virtual care to stretch their workforce

Solving access creates a workforce problem, which creates a budget problem, said Archit Bhatt, Providence’s executive medical director for acute virtual care. Its virtual program covers 78 partner sites across eight service lines, staffed mostly by fractional physicians.

Stanford keeps 28 percent of its ambulatory visits virtual, according to Rosengaus. In southern Oregon, Rick Leesmann, CIO of Sky Lakes Medical Center, serves an area the size of New Hampshire where many patients lack broadband, so his mobile clinics carry tablets and satellite links. Federal rural health transformation funding helps, he said, but it expires.

Governance works when it finds a safe yes and keeps reviewers alert

Two surgeons at TriState Health wore Meta smart glasses during a procedure in October with no policy covering it, and Carpenter shut it down within a day. They had used them in residency and went around her expecting a no. Her team now asks how it can safely get to yes, because clinicians who expect a refusal stop asking and work around governance. The question has moved from can we to “should we, and how,” said Tracy Elmer, chief innovation officer of TrueCare, a Southern California community health center. Attendees also asked whether the data behind these tools represents every patient, and nobody was able to provide an answer.

The room split on whose trust AI has to earn. “The patient isn’t looking to trust an AI agent,” Leesmann said. “They’re looking to trust their caregiver.”

Scott Smitherman, Providence’s CMIO and a practicing internist, said the system does not let AI diagnose or treat, and that it has made him a better doctor. What worries him is the reviewing itself. His ambient notes are now about 99 percent accurate, so he reads them less closely and the rare error slips past. “Human in the loop is absolutely not a panacea,” he said.

Pete Pappas is a journalist and writer covering the intersection of healthcare IT and artificial intelligence. He has written on topics including AI adoption in health systems, ambient clinical documentation, and the practical realities of bringing emerging technology into care delivery. He can be reached at [email protected].