Your Nurses Will Find an AI Tool. The Only Question Is Whose.
Rhonda Dempsey has one rule for any AI tool that comes into a Trilogy Health Services community: it has to take tasks away from the caregivers, and it has to work up front. "If it doesn't work up front, nurses just walk away," she said. That's the bar.
It's a hard bar, and this month two operators cleared it with numbers worth sitting with. Trilogy, which runs more than 140 senior living communities out of Louisville, put a 24-hour chest-worn monitoring device on about 300 high-risk residents, and Dempsey, its chief nursing officer, said it has helped cut hospitalizations by 60%. Signature HealthCARE, 69 locations across five states, expanded fall-prevention AI from four facilities to 20 buildings and reported an 80% decrease in falls where it runs. Both results were operator-reported on a panel at Skilled Nursing News' ReThink conference in Philadelphia, without a stated baseline or time period.
If this sounds familiar, it should. In July we wrote about staff adopting AI before governance caught up. This looks like the same story. It isn't. That one was about the gap. This one is about what closing it looks like, with named operators and reported outcomes.
Here's the part that should get a leader's full attention. Dempsey pointed to the second story underneath the good numbers. "They're going to figure out a way to use some kind of tool to help them," she said of her nurses. If leaders don't give them tools that make the job easier, staff build their own workarounds, and the coverage named where that leads: possible HIPAA violations.
That's a HIPAA exposure sitting inside a success story. It's also predictable, and fixable, if leadership treats it as a speed problem instead of a discipline problem. Nobody needs to convince frontline staff that AI helps. They've already decided that, with their feet. What they need is an approval process fast enough that going around it never seems necessary.
The paper trail is the other half. "I feel like we over document just to make sure we can satisfy every possible auditor that could walk in our building," Dempsey said. Both chief nursing officers made the same point: documentation specialists, with AI's help, are now part of keeping up with auditor demands. That's a staffing line, not a software line. It's also the difference between a defensible record and a discovery exhibit.
An unapproved tool with no documentation trail doesn't fail because the tool was wrong. It fails because nobody can show how the decision got made.
This month, inventory every AI tool actually running in your communities, the ones IT approved and the ones staff may be using on their own. Then name a single owner for the documentation trail behind any AI-assisted clinical decision. Do it before your next survey finds the gap for you.
Read the full research behind this story, plus this week's insurance coverage and governance flagships, in the LITT Briefing on Substack.