Uncorking Care Capacity
What thirty health system leaders taught me about lead time, trust, and the AI that's already in the room
A follow-up to "The Right Blend: From Reactive Staffing to Intentional Workforce Design" (March 19, 2026).
In March, I closed a piece about workforce design with three prompts: where are you muscling through variability instead of redesigning the system, does your labor mix actually flow from your stated strategy, and what one change in the next 90 days would reduce last-minute forcing?
Five months and a few more bottles later, I have an update — and the ground has shifted.
The room has stopped debating whether variability is permanent. That argument is over. What they're arguing about now is lead time and whether they can trust a machine to help them buy it.
The stakes moved in the wrong direction while we were talking about it. The 2026 NSI report put national RN turnover at 17.6% for calendar year 2025 — up 1.2 points, reversing the prior year's improvement — with the average hospital losing roughly $5.19 million a year to RN churn. Every percentage point is worth about $295,000.
On August 6, roughly thirty leaders from more than twenty organizations joined a second virtual wine tasting: CNOs, CEOs and COOs, VPs of strategy and operations, CNIOs, staffing and scheduling directors, nurse managers, perioperative and women's services leaders, home care and post-acute leaders. Systems with dozens of hospitals sat alongside 25-bed rural and behavioral health facilities. Three wines, three questions, no slides, no pitch, nothing on the record.
Here's what came out of the glass.
Wine One: Prosecco — What wakes the palate
Prosecco hits immediately. It's the first thing you notice, and it signals that the evening has begun. Good workforce intelligence should work the same way: something should wake you up while you still have options.
So I asked the simplest question I could. Someone calls out Monday morning. How does that shift actually get filled?
The unit manager or charge nurse texts everyone available and hopes. "Then we cross our fingers."
A central staffing office takes the sick call, broadcasts it, and hands it back to the manager if nobody bites.
A house supervisor reallocates the float pool by greatest need — and if the bench is empty, the manager comes in and works it.
Once a threshold is crossed, the staffing office pages out tiered incentives.
Four systems, four processes, one common ingredient: a human being sending messages into the dark.
Then I asked how far out they'd need to see a gap to do something about it. The answers formed a ladder. Twenty-four hours is enough for one shift. A multi-day absence needs one to two weeks. Six weeks came up over and over in chat as the schedule-building horizon. One leader had tested everything from 24 hours to 10 days and landed on three days as the local sweet spot — not because the math said so, but because human behavior did.
What More Lead Time Makes Possible
The best example of buying lead time came from a large Southeast system. Every Sunday night, a data analyst sends all campus CNOs a rolling three-week staffing view, color-coded red, yellow, and green, with flex pool headcount at the bottom of the page. Leaders can see not just where they're short but whether anyone exists to cover it.
"If several departments are in red, that tells me I'm not getting resources for this campus — so I need to plan differently."
She called it a game changer. Note the technology involved: a spreadsheet, a talented analyst, and a standing Sunday deadline.
Three other mechanics worth stealing:
A behavioral health facility replaced the calling tree with a GroupMe broadcast. The charge nurse posts the need; everyone sees it at once, and staff self-select. It killed the back-and-forth and, more importantly, the guilt dynamic. Staff prefer it. It works because they paired it with heavy PRN hiring, so there's a real pool hunting for hours. Worth sitting with: the most elegant fix described all night was a free group chat, and it worked because it solved a psychological problem, not a math problem.
House supervisors are negotiating across hospitals in a secure chat. One system clusters eight hospitals into regions of three and lets supervisors trade float staff directly. The leader's caveat was the good part — supervisors have to have a real conversation about what staffing actually looks like, because "do you need somebody" gets a different answer than "all my charges have a full assignment and everyone's running seven-to-one."
Functional FTE. Two systems, independently, described building data on current-state leaves plus historical call-in run rates — which trend remarkably consistently by unit — to justify hiring above budgeted FTE, with finance and HR at the table, and feeding it back into position control.
That last one is the economics nobody codifies, finally codified.
And one warning that reframes the whole conversation. A system float leader described building six-week assignment blocks, then watching units cancel those staff when the need evaporated between schedule build and shift date. Which means lead time isn't only a coverage problem. Get the look-ahead wrong, and you burn the internal bench you built specifically to avoid agency.
How Far Can You Reach Before Going External?
Wine Two: Bordeaux Blanc — What's underneath
The Graves Blanc looks like a simple white. Underneath, the Sémillon is doing the structural work you don't see. Staffing is the same: the surface is a schedule; the substance is everything that feeds it.
In March, AI got a passing mention. This time it dominated — which tracks with the market. A February 2026 survey of 120 health systems found that 75% are now using or planning to use at least one AI application, up from 59% a year earlier, and that 67% of systems are running three or more.
But the interesting part in our room wasn't capability. It was trust.
One system runs AI-assisted predictive staffing across women's health inpatient units and EDs, layered on top of UKG. It pushes the six-week schedule, provides a seven-day look-ahead that factors in incentive policy and historical callout patterns, flags gaps ("this person hasn't met their FTEs, you're missing a charge nurse"), and supports workforce planning against seasonality and turnover so they can pre-hire ahead of known vacancy cycles. What it explicitly does not do is assign anyone. "There's no auto part of it." The leader telling me this was drinking water, because she'd been up since 4:00 am rounding at a hospital with angst about the change.
A workforce leader at a large multi-state academic system described the most mature build I heard that night: AI predicting both patient demand and clinical workload, feeding FTE forecasting, which in turn drives schedule prediction. Demand, workload, staffing plan — one chain instead of three disconnected exercises.
I asked how they built trust in it. The sequence matters more than the technology: finance and executive nursing leadership first, operational leaders second. Then they ran a period showing the manual outcomes alongside the system's recommendations, so everyone could see for themselves where the gaps actually were.
Then I asked whether accuracy or explainability mattered more. Both, she said — but neither was the thing she'd rank first. What mattered most was a visible commitment that humans stay in the loop running daily operations.
Another leader gave that its teeth:
"The AI can tell us the number. The human factor knows who's having challenges at home. We know the personal lives of our employees."
That's the whole argument for human-in-the-loop in two sentences, and it came from an operator, not a vendor.
AI Should Buy Time — Not Make the Decision
In the room, the adoption gap ran in one direction. Clinical AI was already in production — drafting patient-message replies through clinical pools, routing documents into the record, and predicting clinic volumes and patient access. On the workforce side, leaders still described spreadsheets and text messages. One was blunt: resume screening still isn't good enough for certain roles.
Meanwhile, a national system is piloting 24/7 AI voice screening for CNA candidates. Recruiters work nine to five, so anyone coming off a night shift was effectively unreachable. Worse, candidates often heard back only after a role had been filled, leaving them to assume they had been ignored. Her phrase for the fix: it leaves no candidate behind. Recruiters and hiring leaders still make the decisions. And as administrative work gets absorbed, they are telling recruiters this is the moment to build the relationship skills that were always the actual job.
Hold that thought. It comes back in the third glass.
One more lever I didn't expect came from the ambulatory side of the room.
When leaders can see a site is likely to be short, they do not only go hunting for coverage. They change the visit. Certain clinics have been deliberately tailored to switch to a telemedicine model — and in that model, the MA, RN, and other support staff are not needed physically on-site.
That inverts the whole exercise. Every other mechanic in this article moves people toward demand. This one reduces the demand for on-site people. In her words, it “allows you to think a little differently about what you're sourcing, what's high risk, and pivoting where you can.”
The prerequisite is the part worth stealing: those clinics were designed to be switchable before anyone needed to switch them. That is not a same-day improvisation; it is a care-model choice made in advance.
The same ambulatory leader described using AI to forecast clinic volumes and patient access. Same visibility, different response: redesign the visit instead of scrambling to staff it.
Wine Three: Pinot Noir — What design buys you
Pinot Noir is unforgiving. You can't muscle it. The result comes from site, choices, and restraint — which is a decent description of a workforce model.
So, think about the nurses who stayed for years. What made them stay?
Convenience. The commute. Getting the shift they wanted. Their coworkers. Culture. Growth — but only real growth, not onboarding language. And then the line of the night:
"The reason people stayed then is not why they stay now. Flexibility is the new currency."
Which is also, everyone agreed, the hardest thing to deliver consistently—partly because it means different things to different people. Some want a say in their schedule. Others want mobility. One leader described an employee cross-trained across specialties who says she's two different people—her ICU self and her ED self—and rotates so she doesn't burn out in either.
Three retention levers this room named that don't show up in most engagement surveys:
Contract labor is a dissatisfier. Being fully staffed doesn't help if a slice of that staffing consists of travelers working alongside your permanent staff. This is the part the cascade chart can't show: reaching to the bottom of the cascade puts pressure on the top of it. Fully staffed on the spreadsheet and quietly eroding on the unit is a real state, and plenty of systems are in it.
Highlighting expertise travels further than you think — including to agency staff. One leader spent part of that same day calling unit leaders to ask whom they'd recommend by name for a unique need.
Virtual nursing keeps seasoned staff. Move experienced nurses off twelve-hour bedside shifts into virtual roles for admissions and blood transfusion checks, and they stay — and mentor. Worth being precise here: the published evidence on virtual nursing's workload effects is genuinely mixed and depends heavily on local implementation. The retention case — role diversification that keeps experienced clinicians in the building — is the stronger one, and it's the one this room was making.
Notice that virtual nursing and the AI voice agent are the same move. Both strip the administrative tax off a role so the human can do the relationship work. That's the through-line between the second and third glasses.
And that last lever is the whole thesis in miniature. It doesn't add headcount. It redesigns where existing capacity sits.
What I'm taking away
The future of workforce strategy isn't about filling holes faster. It's better visibility, better options, and enough time to make a thoughtful decision. The goal isn't to control every surge — it's to design a model that moves with the current instead of fighting it.
Force versus flow, still.
Three questions to take back to your leadership team:
How much lead time does your model actually give a frontline leader — not in theory, in practice, this Monday?
How far down your internal bench can you reach before you touch external labor, and can you prove the cost of every tier you skipped?
Where are you asking people to trust a recommendation without showing them the manual comparison?
If you're working on any of this — the three-week glance, functional FTE, the cascade economics, the trust sequence — I'd like to compare notes. Reply here or reach out directly.
My thanks to everyone who joined; to Haley Moore for guiding us through the wines; and to Medely for convening a room with no slides, no demo, and no agenda beyond a candid conversation.
Greg Paulus is the founder of WonWay Health, a consultancy focused on healthcare workforce innovation. He has spent 30 years in healthcare, with the last 20 focused on health and workforce technology.
Comments