The Agency Line Test. Every AI Capacity Figure Is Counted in Hours. Your Cost Report Says Whether They Convert.

A vendor will walk into your building this quarter with a number measured in hours. Minutes returned per shift. Discharge hours absorbed. Capacity described as the equivalent of so many nursing positions. Most of those figures are real and most of them are published.

None of them is money yet.

Which line on your budget absorbs those hours is a question about your hospital rather than about the technology, and you can answer it before the meeting starts.

▪Every AI capacity figure published this year is counted in hours, and hours are not dollars

Start with what the published cases actually report. Becker’s compilation of virtual nursing results across ten systems, December 2025, is the cleanest list available. Piedmont reports 84,460 discharge hours handled virtually. WellSpan reports 4,300 hours across eight hospitals, running at roughly two hours a shift. Essentia covers 22 units with six virtual nurses and reports more than 2,100 hours.

Every one of those is an hours figure. Hours are capacity, and capacity converts into cash in exactly three places.

You stop buying contract labor. You stop paying overtime and premium shift rates. Or you do not open a requisition you would otherwise have opened.

The three are not interchangeable, and only the first two show up as money in the fiscal year you are defending. The third shows up as a line that never posts, which no budget records as a saving.

A published time saving is not a cancelled requisition, and anyone presenting it as one will lose the room the first time a CFO asks which account it comes out of.

▪Your contract-labor line is the only one of the three a finance committee can verify

Set the scale first. Labor is 56% of what a hospital spends, about $890 billion in 2024 by the American Hospital Association’s accounting. Contract labor is the slice of that you are renting rather than employing, and it is the slice a capacity program can actually cancel.

Here is the part you can look up rather than argue about. We pulled contract labor for direct patient care and total salaries from the CMS Hospital Cost Report, took one filing per hospital — the most recent complete fiscal year, ending in 2024 or 2025 — and kept the 2,901 hospitals that report both.

Together they spent $20.8 billion on contract labor against $436.6 billion in salaries. That is 4.77% in aggregate.

The median hospital sits at 3.55%. The first quartile is at 1.46% and the third at 7.23%. The ninetieth percentile is 12.53%.

Those four numbers are the whole business case. 432 of the 2,901 — one hospital in seven — spend 10% or more of their salary line on contract labor, and those hospitals hold 38.3% of every contract-labor dollar in the file, averaging $18.5 million each. At the other end, 939 hospitals, close to a third, spend under 2%, and together they account for just 6.6% of the dollars.

If you are in the first group, hours returned have somewhere to go and a comptroller can watch them go there. If you are in the second, the agency line is already gone, and the hours have to be argued on a slower and softer proof.

▪The size of your hospital tells you nothing about which group you are in

Now put the obvious predictor beside it. Hospital size does not move this number at all.

Median contract-labor share runs 3.47% under 100 beds, 3.58% at 100 to 199, 3.66% at 200 to 299, 3.55% at 300 to 499 and 3.60% at 500 and above. The share of hospitals above the 10% threshold runs between 13.4% and 15.9% in every one of those bands. Rural and urban barely separate either, at a 3.68% and 3.46% median.

The 600-bed system with the capital budget for a virtual nursing platform is no more likely to have agency spend available to cancel than the 80-bed hospital that cannot fund one.

Geography does move it. Among states with at least 25 reporting hospitals, Puerto Rico’s median is 9.65%, Iowa’s 7.95%, Washington’s 6.04%, West Virginia’s 6.00%, Arizona’s 5.93% and Massachusetts’ 5.74%. Inside a single state the spread is wider than the gap between most states — a quarter of Arizona’s reporting hospitals sit above 10% while its median is under 6%. Our contract-labor benchmarks in the Workforce Intelligence Center carry this state by state and refresh weekly, if you want your own market rather than the national figure.

▪What the hours do when there is no agency line left to cancel

For the third of hospitals already under 2%, the capacity still arrives. It lands on turnover and on reqs that quietly do not reopen, and the published cases do carry numbers there.

Piedmont reports 85% of its nurses saying they like working with the virtual team and 78% saying they are more likely to stay. WellSpan reports an 8% decrease in turnover at one hospital. CoxHealth reported turnover falling from 25% to 11% between June 2023 and Becker’s compilation of November 2024, which is now a two-year-old figure and should be quoted with its date.

Run that against your own replacement cost. NSI’s 2026 report puts the average cost of a single RN separation at $60,090. A 400-RN hospital at the 17.6% national turnover rate loses roughly 70 RNs a year; a three-point improvement is about twelve fewer separations, or $721,000. That is The RN Network’s own arithmetic from NSI’s figures, and it moves entirely with your own cost per separation.

My read: the next year of AI procurement in nursing gets argued on the contract-labor line, because it is the only one of the three that survives an audit — and the hospitals already under 2% will be sold the same case with no account left to take it from.

▪Five moves before the next AI business case reaches your CFO

Look up your own number first. Contract labor for direct patient care, divided by total salaries, from your most recent cost report. Put the percentage on page one of the business case and let it set which argument you are making.

Make the vendor name the line. Ask in writing which budget line the published hours came off at the reference site, and whether that site had contract labor to cancel. A result from a 12% hospital does not transfer to a 1.5% hospital, and the deck will not say so.

Price your own pitch to the account’s number. For recruiters and staffing firms, an account under 2% is not an agency-displacement conversation. Time to fill and first-year retention are what is actually in play there.

Know which of your accounts sit in the tail. 38% of contract-labor dollars sit in one hospital in seven. Those are the accounts where a capacity program is aimed at your spend, and they are not the ones you would pick by bed count.

Convert hours into a range, never a point estimate, in your own headcount plan. Give your finance partner a floor and a ceiling with the assumption attached. A single number you cannot defend becomes the number you are held to.

▪Nobody gets a line item for capacity that arrived without a hire

Every hiring leader I talk to is measured on time to fill, cost per hire and offer acceptance. Not one of them is measured on hours a virtual nurse absorbed on a med-surg unit, and that is exactly why the capacity argument keeps getting made somewhere else and then handed to talent acquisition as a headcount target.

Get in front of it with the one number that decides whether any of it converts. Pull your contract-labor share this week, before the business case lands on your desk with a savings figure already written into it.