Care operations and capacity
A forecast frees no beds on its own. Capacity AI pays off when each prediction lands with a named team that has the authority and the hours to act on it the same day.
Part of AI for Life Sciences & Healthcare. 7 sources, checked .
- mean US hospital occupancy in the year after the COVID-19 emergency ended (May 2023 to April 2024), up from 63.9% in 2009 to 2019
- 75.3%
- mean US hospital occupancy in the year after the COVID-19 emergency ended (May 2023 to April 2024), up from 63.9% in 2009 to 2019
- JAMA Network Open, 2025(opens in a new tab)
- drop in staffed US hospital beds from pre-pandemic levels, the main driver of higher occupancy
- 16%
- drop in staffed US hospital beds from pre-pandemic levels, the main driver of higher occupancy
- JAMA Network Open, 2025(opens in a new tab)
- spent by US hospitals on their workforce in 2025, their largest expense
- $1tn+
- spent by US hospitals on their workforce in 2025, their largest expense
- American Hospital Association, 2026(opens in a new tab)
The problem
US hospitals run fuller than they did before the pandemic. Mean occupancy rose from 63.9% across 2009 to 2019 to 75.3% in the year after the COVID-19 public health emergency ended.1 Supply drove it, not demand: staffed beds fell from about 802,000 to 674,000 while mean daily census held near 510,000.1 If nothing changes, national occupancy could reach 85%, a level experts treat as a bed shortage, by about 2032 for adult beds.1
Staff is both the binding constraint and the largest cost. Hospitals spent more than $1 trillion on their workforce in 2025, and workforce costs rose 5.6% that year while inpatient volumes grew 5.3%.2 Every staffed bed depends on nurses scheduled against real demand, and Medicare's Conditions of Participation require adequate numbers of nurses and other personnel to care for all patients as needed.3
Crowding now has a federal yardstick. CMS adopted the Emergency Care Access & Timeliness eCQM, which tracks waits over one hour, patients who leave unseen, boarding over four hours and ED stays over eight hours, with mandatory reporting from 2028.4 CMS notes that boarding often occurs because of shortages of inpatient beds and staff, and cites evidence that each boarded patient adds at least 12 minutes to median ED length of stay for admitted patients.4
Operational AI is spreading fast. Among hospitals using predictive AI, use for scheduling rose from 51% to 67% between 2023 and 2024.5 Outcome evidence is thinner but instructive. In a prospective implementation in South Australia, a discharge-likelihood model was handed to a weekend team of a senior nurse and a pharmacist, and median length of stay fell from 3.1 to 2.9 days with lower seven-day readmissions.6 The model came with a team whose job was to act on it.
What we build
- 01
Admission and census forecasting
Forecasts of ED arrivals, admissions and discharges by unit, refreshed through the day and fed into the bed huddle with stated error ranges.
- 02
Discharge readiness lists
A daily ranked list of patients likely to be ready for discharge within 24 hours, routed to case managers with the open barrier for each patient, such as transport, pharmacy or placement.
- 03
Demand-matched nurse staffing
Shift-level staffing recommendations from forecast census and acuity, built into your scheduling system so managers adjust before the shift starts, not during it.
- 04
Operating room case-length prediction
Case duration estimates from procedure, surgeon and patient history, used to reduce overruns and release unused block time early.
- 05
Boarding early warning
Alerts when projected boarding will breach the four-hour threshold, listing the inpatient units that can release beds and the discharges that would free them.
Where it goes wrong
The failure modes we design against from the first week, and the control for each.
Forecasts nobody owns
A dashboard without an owner changes nothing. Control: assign each prediction to a role with authority to act, such as a flow or discharge team, and track the actions taken. The South Australian result came from a model paired with a dedicated team.6
Discharges pushed before they are safe
Optimizing for length of stay can move patients out before post-hospital care is arranged. Hospitals must identify at-risk patients early and plan discharge in time to arrange that care.7 Control: track readmissions alongside length of stay and keep the discharge decision with clinicians.
Staffing models that learn past shortfalls
A model trained on historical rosters learns the staffing you had, not the staffing patients needed. Control: train on demand and acuity, test recommendations against the adequate-staffing standard in the Conditions of Participation,3 and review them with nurse leaders.
Forecasts that break when patterns change
Respiratory seasons, new service lines and nearby closures shift demand. Control: publish an error range with every forecast, monitor accuracy weekly, and retrain on a schedule with human sign-off.
Rules that apply
CMS Emergency Care Access & Timeliness eCQM (CY 2026 OPPS final rule)
Primary text on federalregister.gov(opens in a new tab)Hospitals report the share of ED visits with waits over one hour, patients leaving unseen, boarding over four hours and stays over eight hours, voluntarily for 2027 and mandatorily from 2028.
Medicare Conditions of Participation, nursing services, 42 CFR 482.23(b)
Primary text on ecfr.gov(opens in a new tab)The nursing service must have adequate numbers of nurses and other personnel to care for all patients as needed, so staffing models are judged against patient need, not budget alone.
Medicare Conditions of Participation, discharge planning, 42 CFR 482.43
Primary text on ecfr.gov(opens in a new tab)Hospitals must identify patients at risk early and plan discharge in time to arrange post-hospital care, so discharge prediction supports this process rather than replacing it.
How to measure it
- Forecast error for daily admissions and discharges, by unit
- ED boarding hours and the share of admitted patients boarded over four hours
- Median length of stay alongside 7-day and 30-day readmission rates
- Time from discharge order to departure
- Overtime and agency hours as a share of total nursing hours
Questions buyers ask
Can AI reduce emergency department boarding?
It can help when boarding comes from inpatient flow, which CMS identifies as a common cause. Forecasts of admissions and discharges let bed managers open capacity before the ED backs up. The gain comes from earlier discharge work and staffing changes, not from the forecast itself. CMS's new emergency care measure will make boarding hours visible, so set your baseline now.
What data do we need for bed and staffing forecasts?
Usually data you already hold: admission, discharge and transfer events, ED tracking timestamps, OR schedules, staffing rosters and acuity scores. Two to three years of history covers seasonal patterns. The hard part is quality, because missing discharge times and late documentation distort forecasts. We fix the data feeds before we tune any model.
Will AI staffing recommendations replace nurse managers' judgment?
No, and they should not. Recommendations arrive before the shift with the forecast and its uncertainty, and the charge nurse or manager decides. Managers know about call-outs, skill mix and patient needs the model cannot see. Track how often recommendations are accepted and why they are overridden, because that feedback is what improves the model.
Sources
Every figure on this page comes from the sources below. We checked each link on .
- 1.Health Care Staffing Shortages and Potential National Hospital Bed Shortage(opens in a new tab)JAMA Network Open, February 3, 2025 · pmc.ncbi.nlm.nih.gov
- 2.Costs of Caring: Challenges Facing America's Hospitals as They Care for Patients in 2026(opens in a new tab)American Hospital Association, March 2026 · aha.org
- 3.42 CFR 482.23: Condition of participation: Nursing services(opens in a new tab)Electronic Code of Federal Regulations (CMS) · ecfr.gov
- 4.Medicare Program: Hospital Outpatient Prospective Payment and Ambulatory Surgical Center Payment Systems; Quality Reporting Programs (CY 2026 final rule, 90 FR 53448)(opens in a new tab)Centers for Medicare & Medicaid Services (Federal Register), November 25, 2025 · federalregister.gov
- 5.Hospital Trends in the Use, Evaluation, and Governance of Predictive AI, 2023-2024 (ASTP Data Brief No. 80)(opens in a new tab)Assistant Secretary for Technology Policy / Office of the National Coordinator for Health IT, HHS, September 2025 · healthit.gov
- 6.The Adelaide Score: prospective implementation of an artificial intelligence system to improve hospital and cost efficiency(opens in a new tab)ANZ Journal of Surgery, January 3, 2025 · pubmed.ncbi.nlm.nih.gov
- 7.42 CFR 482.43: Condition of participation: Discharge planning(opens in a new tab)Electronic Code of Federal Regulations (CMS) · ecfr.gov