Adoption Outran Accounting
Walk onto any floor and ask the nurse manager how many AI tools she has running and who owns each one. Watch her count. One arrived with a corporate initiative. One came from a pilot that quietly became permanent. One rode in on a vendor bundle nobody unbundled. Now ask who owns the third. The answer you will get: somebody at corporate, probably.
That conversation now has data behind it. The eighth annual McKnight's Mood of the Market poll, fielded online from June 23 through August 11 among 578 nursing home owners, C-suite executives, administrators and nursing managers, found that 27.9% now say AI is already useful to their organization, up from 17.1% a year earlier. The share calling AI not really needed or applicable fell from 15.7% to 8.5% (McKnight's Long-Term Care News). The resistance argument is over. Leaders bought in.
In the same survey, 25.8% of respondents reported no actual savings in any of the twelve automation or technology categories offered. Among nurses that number was 32.4%. And 38.9% said their organization is still doing too little or way too little with AI.
Read those findings together and you get the year's most uncomfortable sentence. A meaningful number of operators are convinced they are behind, and cannot demonstrate that what they already bought worked.
A separate survey published the same month names the mechanism. The 2026 CTO Hotline report from LeadingAge's Center for Aging Services Technologies and Ziegler Link-age Funds, drawn from technology leaders at nonprofit senior living and care organizations, found AI capability inside those organizations jumped from 13% of respondents in 2025 to 55% in 2026. It also found that 82% of respondents set success metrics when they implement new technology, and only 39% usually go back to those metrics to assess results (LeadingAge Ohio). Different sample, different population, same answer from the opposite direction.
The metrics get written. Almost nobody returns to read them.
The sector absorbed an average 2.4% Medicare rate increase this year. There is no margin left to fund a second round of unmeasured pilots.
Look at where savings actually landed and the picture sharpens. Clinical and documentation tools were the number one category, named by 32.4% of respondents. Predictive analytics and business intelligence came in at 8.7%. Wander management and elopement tools came in at 8.5%. The categories that dominate conference agendas sit near the bottom of the realized-savings list. The unglamorous one sits at the top.
Then there is the finding almost nobody quotes. Asked which physical or operational improvement would most improve job satisfaction and efficiency, 41.9% of all respondents chose communication and connectivity infrastructure, including reliable Wi-Fi, staff coordination apps and mobile devices. Among administrators it was 46.2%. Among nurse leaders, 43.3%. In a survey about AI enthusiasm, the top-ranked intervention was Wi-Fi.
The operators posting results are the ones who took that seriously. United Methodist Communities runs two dedicated gigabit circuits per community and swapped medication cart laptops that lasted half a shift for Chromebooks that last two days, per Travis Gleinig, its vice president of innovation and CIO (HealthTech Magazine). Asbury Communities tests roughly five technologies at a time for 60 to 90 days each in an innovation lab, decided its fall detection cameras would generate visual representations rather than identifiable images before the sales conversation, and got 98% of assisted living and skilled nursing residents to opt in, according to Chief Operating Officer Todd Andrews. And 2Life Communities, which houses more than 2,000 older adults around Boston, trained 102 residents one at a time on an AI scam-verification tool, working through a resident technology support program that had already earned their trust. Residents submitted 682 verifications, and the tool flagged 204 scams and 184 suspicious messages (LeadingAge CAST).
None of those are technology decisions. Every one of them is a sequencing decision made before the purchase order.
Here is the action for this month, and it costs nothing but honesty. Pull your technology spend for the last four quarters and sort it into the same twelve categories the survey used. Beside each line, write one sentence naming the metric that proves it worked. Then add three columns to your AI inventory that most inventories do not have: the baseline captured before go-live, the savings realized to date, and the liability cap in the vendor contract. Whatever comes back blank is your finding. Fix the connectivity layer before you fund another pilot. And since 82% of your peers already set metrics while 39% return to them, write the house rule to cover the half that fails. No AI tool goes live without a named owner, a measured baseline, and a review date already on the operations calendar with a person's name against it.