Industries / Health / Healthcare / Nurse staffing
Scheduling
Scheduling is how a hospital fills every shift on every ward: first with its own permanent nurses, then with its own bank, float or per-diem nurses, and only then with nurses borrowed from agencies.
7
Milestones achieved, from 1976 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1976
Software weighs nurses' wishes
D. Michael Warner described a nurse scheduling system that treats a four- or six-week schedule as one large optimization problem. Every schedule it considers puts the minimum number of nurses of each skill on each day and shift, and among those it picks the one that best fits each nurse's stated wishes on length of work stretches, rotation and days off. The paper describes its use on nursing units of several hospitals.
Credit
D. Michael Warner (Operations Research, October 1976)
2001
England starts a national bank
The Department of Health set up NHS Professionals in 2001, a temporary staffing service run by the NHS itself, to cut the cost and raise the quality of temporary nurses. A House of Commons committee report of the 2006–07 session said that by 2006 it was used by only 24% of trusts, and that the department accepted it had "a troubled history" of managerial and financial trouble during rapid expansion.
LessonA national staff bank changes little unless hospitals choose it over their own banks and agencies.
Credit
Department of Health; NHS Professionals
2013
Hospitals build in-house agencies
Henry Ford Health System in Detroit started an internal staffing unit in 2013: its nurses earn more per hour than regular staff and choose their shifts, and its director says it costs much less than outside agency staff. During COVID-19 others followed. Piedmont Healthcare in Atlanta, whose chief executive said its agency nurse spending rose from $20 million a year before the pandemic to $400 million in one year, about a third of it kept by the agencies, formed its own staffing unit. AMN Healthcare's 2025 annual report lists hospital systems with "internal travel agencies" among its competitors.
Credit
Henry Ford Health System; Piedmont Healthcare
2013
Agency nurses and deaths
Linda Aiken and colleagues linked a 2006 survey of 40,356 registered nurses in 665 hospitals in four US states with outcomes for about 1.3 million patients. Hospitals that used more nurses employed by agencies at first seemed to have higher death rates, but the link disappeared once the quality of each hospital's work environment was taken into account.
LessonHospitals that lean on agency nurses often have poor working conditions, and the conditions, not the agency nurses, explained the extra deaths.
Credit
Linda Aiken, Jingjing Shang, Ying Xue and Douglas Sloane (Health Services Research, 2013)
2016
One measure for ward staffing
Lord Carter of Coles' review of 32 English hospital trusts created Care Hours per Patient Day (CHPPD): the hours worked by registered nurses and support workers on a ward, divided by the number of patient days. Data from more than 1,000 wards ranged from 6.3 to 15.48 CHPPD. The review found that most trusts had electronic rostering but few used it fully, and recommended that every trust use it, publish rosters six weeks ahead, and report CHPPD monthly from April 2016.
Credit
Lord Carter of Coles, for the Department of Health (February 2016)
2022
Acuity staffing proves slow
New Zealand's Care Capacity Demand Management (CCDM) programme, set up after a 2006 inquiry, uses TrendCare software to measure how much care each patient needs and match nurse numbers to it. After the 2018 national nurses' strike, a Safe Staffing Accord required every district health board to put it in place by 30 June 2021. A review for the Minister of Health in February 2022 found it unevenly implemented, with no agreed meaning of "implemented", serious data-quality problems and limited evidence of better patient outcomes; 62% of frontline nurses surveyed disagreed or strongly disagreed that TrendCare had "a positive impact on patient care".
LessonA tool that matches staff to patient need works only when the data put into it are complete and trusted.
Credit
Nursing Advisory Group, for New Zealand's Minister of Health
2023
Borrowed shifts count for less
Billy Zaranko and colleagues linked rosters with 66,923 admissions on 53 wards at three hospitals of one English NHS trust in 2017. Each extra 12-hour shift by a registered nurse employed by the hospital went with 9.6% lower odds of a patient dying; no link was found for agency nurses or support workers, whom the authors judged "not effective substitutes". A 2025 study of 70 US hospitals from 2019 to 2022, by Patricia Pittman and colleagues, linked agency nurse hours above a threshold with more pressure ulcers (bed sores), a measure especially sensitive to nursing care.
Credit
Billy Zaranko, Anne Marie Rafferty and colleagues (BMJ Quality & Safety, 2023); Patricia Pittman and colleagues (JAMA Network Open, 2025)
Watch for
Text agents fill gaps
Chromie Health says Chromie Dispatch, an AI agent that texts a hospital's own qualified nurses when a shift opens and books those who reply yes, fills gaps 60 times faster than manual scheduling and has covered more than 35,000 last-minute shifts. A nurse manager at St Joseph's Hospital is quoted as saying her unit's overtime costs fell 25%.
Achieved whenan independent evaluation at more than one hospital measures time to fill, shifts left open and cost against the hospital's earlier method.
Claimed by
Chromie Health, May 2026
An AI float pool trial
The University of Hawaii registered a trial in May 2025 of GAINS, a generative AI staffing method for the medical-surgical float pool at The Queen's Medical Center, later extended to the nursing assistant pool. It will measure burnout, planned against actual staffing, and staffing costs. At its last update it was not yet recruiting.
Achieved whenthe trial publishes staffing and cost results against the period before the change.
Claimed by
Katie Azama, University of Hawaii, May 2025
Open
01
Float pool sizing tool
An open-source simulator that takes a hospital's daily inpatient census, a staffing target in nurse hours per patient day, and the hourly cost of each kind of cover (permanent staff, overtime, a float pool, per-diem nurses, internal travel nurses and agency nurses), and reports the float pool size and mix that keep staffing shortfalls under a set level at the least cost. Run it on public census series for at least 50 US hospitals.
The briefCredit · unclaimed02
Shift-offer agents under test
A staged test of AI agents that text a hospital's own staff to fill open shifts. Volunteers play 50 made-up nurses, each following a script of availability, hours already worked, unit skills and preferences. Each agent must fill 30 open shifts in the order the hospital sets (float pool, then per-diem, then overtime, and only then an agency request), without offering a shift to anyone who lacks the unit's skills, would pass a weekly hour limit or would get less than the required rest between shifts.
The briefCredit · unclaimed
Open 01 · the brief
Float pool sizing tool
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open-source simulator that takes a hospital's daily inpatient census, a staffing target in nurse hours per patient day, and the hourly cost of each kind of cover (permanent staff, overtime, a float pool, per-diem nurses, internal travel nurses and agency nurses), and reports the float pool size and mix that keep staffing shortfalls under a set level at the least cost. Run it on public census series for at least 50 US hospitals.
Why it is still open
Published models size flexible staff for one unit with one hospital's data. Yue Hu, Carri Chan and Jing Dong's 2024 study of an emergency department estimated 10% to 16% lower staffing costs in simulation, and a 2026 study simulated a single 20-bed intensive care unit from MIMIC-IV records. A GitHub search in November 2026 found no general tool, only one empty repository.
Built with what exists
The US health department's facility-level dataset of weekly inpatient beds in use at each reporting hospital from 2020 to May 2024 (healthdata.gov dataset anag-cw7u); SimPy, a Python library for simulation; Google OR-Tools for the cost minimization; hourly costs set by the user.
First step
Download two years of weekly census for five hospitals of different sizes, and write down and code a method for turning weekly averages into daily series with realistic day-to-day swings.
Done looks like
On 20 made-up hospitals whose best float pool size is found by brute force, the tool finds that size every time; for 50 real hospitals' public census series it publishes the least-cost mix and how it changes when agency rates rise 50%.
Open 02 · the brief
Shift-offer agents under test
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A staged test of AI agents that text a hospital's own staff to fill open shifts. Volunteers play 50 made-up nurses, each following a script of availability, hours already worked, unit skills and preferences. Each agent must fill 30 open shifts in the order the hospital sets (float pool, then per-diem, then overtime, and only then an agency request), without offering a shift to anyone who lacks the unit's skills, would pass a weekly hour limit or would get less than the required rest between shifts.
Why it is still open
Hospitals still fill last-minute gaps "through phone calls, group texts, and spreadsheets", by Chromie Health's account, and the text agents now sold for the job, Chromie Dispatch and vflok's Swift (a company begun by Catalyst By Wellstar, the Wellstar health system's venture arm, and piloted in three hospitals from January to June 2025), are known only from company and partner accounts with no outside test.
Built with what exists
Twilio for text messages; agents built on current models such as Claude or Meta's Muse Ember (see the Note "The jobs agents repeat"); Google OR-Tools CP-SAT to compute the best possible assignment for each scenario as the answer key; a scoring script.
First step
Write five scenarios with a nurse manager, including a 5 a.m. sick call and a sudden rise in census, and run them with volunteers against one agent.
Done looks like
At least 3 agents and one fixed-rule script each run all 30 scenarios; published for each: share of shifts filled, rule breaks per 100 offers, minutes to fill, messages per filled shift, and how evenly offers were spread across staff.