Ambient artificial intelligence handed many physicians their evenings back. For nurses, the same promise is proving harder to keep, and the reason is buried in the flow sheet.
Physician ambient scribes listen to a visit and draft a narrative note when it ends. Nursing documentation does not work like that. It lives in structured flow sheets, entered field by field across a shift, every time a nurse assesses, medicates or turns a patient.
“Nursing documentation occurs throughout the entire shift. It’s every time we interact with a patient,” said Stephanie Clements, BSN, RN, senior vice president and chief nurse executive at Chesterfield, Mo.-based Mercy. “It’s very different than summarizing care at the end of a period.”
That difference is why early attempts to copy the physician model — record the encounter, generate the note — have not mapped cleanly onto the bedside, and why some nurses have found the tools slower, not faster, at first.
Yet the systems furthest along say the payoff is real when the technology is built for nurses from the get-go.
At Charlotte, N.C.-based Advocate Health, which rolled out Microsoft’s ambient documentation to nursing units, the gap between a nurse assessing a patient and that assessment reaching the record — what Advocate calls documentation latency — fell from 52 minutes to 17 over 13 weeks, according to Tracy Breece, MSN, RN, vice president of nursing innovation, AI and emerging technologies. Nurses have saved more than 94,000 clicks since the scale-up, she said, and 80% report meaningful time savings.
“The goal is never really about the clicks for me, and it’s not about making documentation faster,” Ms. Breece said. The restored time, she said, is what lets a nurse have a break, a full lunch break or simply leave work on time.
Since implementing the tool, the 69-hospital system has filed more than 61,000 ambient recordings and more than 1 million nursing workflow sheets supported by ambient AI.
Mercy, which says it is the first health system to deploy the Android version of the tool, measured a 22% drop in the time nurses spend in flow-sheet documentation across three pilot units, as well as a 29% to 56% reduction in overtime.
The catch is that none of it is really about the software.
“I think it’s going to be a total culture change,” said Tiffany Murdock, DNP, RN, senior vice president and chief nursing and clinical excellence officer at Ochsner Health, of ambient listening in nursing. “We are going to be so awkward and uncomfortable actually talking to our patients about the assessment that we do. It’s going to be really weird for us to do it.”
Earlier this year, Colleen Mallozzi, RN, senior vice president and chief nursing informatics officer at Philadelphia-based Jefferson Health, told Becker’s there is “an awkwardness to get over” for nurses using ambient AI.
Nurses often complete a head-to-toe assessment in silence, Dr. Murdock said. An ambient tool only works if they say aloud what they are observing, so the microphone can capture it.
The mechanics turn delicate fast. Dr. Murdock described a nurse noting that a patient is tachypneic — breathing quickly — only to have the patient object: “No, I’m not breathing fast. That’s how I normally breathe.” A mature ambient model, she said, has to edit the room’s cross-talk down to the clinically accurate entry.
“There’s some things you have to work out. But I think it’s more of a change management than an actual technology [problem],” Dr. Murdock said.
Advocate reached the same conclusion and stopped treating the rollout as a software launch.
“We viewed ambient documentation as a practice transformation. This was not a technology deployment,” Ms. Breece said.
The system built an avatar-based simulation training program where nurses rehearse “care out loud” before they walk into a patient room; 67% who practiced with the avatar reported more confidence speaking their assessments aloud.
When a nurse tells a manager the new workflow is slower, Ms. Breece said, “in the beginning, that’s a true statement,” because the team is unlearning years of muscle memory.
The leaders often returned to a blunter warning: Do not automate a broken process. Dr. Murdock recalled an effort to use remote patient monitoring to spare nurses manual vital-sign checks that Ochsner nearly “reverse engineered” — and that ended up helping nurse techs more than the nurses it was designed for.
“It cannot be put on top of a poor process,” she said. “You’re not trying to fix the process through the technology, because it just becomes more clunky.”
What separates the systems seeing returns is what they choose to measure. Advocate tracks confidence, not compliance — 98% of the more than 100 nurse leaders who completed a three-hour change-management workshop said they felt more confident leading the shift. Mercy watches which nurses adopt fastest and turns them into unit champions, who “self-identify,” Ms. Clements said, and prove more persuasive than any informaticist. Its most surprising converts have been veteran nurses: “Oh my goodness, I’ve been waiting for this. Where has this been my whole career?”
Both systems are already pushing the change upstream, working with nursing schools so students practice narrating care — Mercy calls its training sandbox the “playground” — before they reach a real patient. The endpoint, Dr. Murdock said, is not a faster keyboard but no keyboard at all: “any technology that brings the human back to the bedside.”
