Healthcare AI is often discussed as a diagnostic technology that works at the level of one patient. A project from the University of Hertfordshire and regional NHS bodies addresses a different question: can machine learning help an entire health system prepare for demand before beds, staff, or services come under pressure?
The operational forecasting model uses five years of historical information, including admissions, treatments, readmissions, bed capacity, infrastructure pressure, and workforce availability. It also accounts for local demographics such as age, gender, ethnicity, and deprivation.
Researchers intend the system to show how demand may change across short, medium, and long horizons. Testing is underway in hospitals, with community services and care homes planned as later environments. The work is due to continue through 2026.
A Forecast Makes Inaction Visible
Public organisations often possess large historical archives that explain what happened without changing what happens next. Forecasting can turn those records into scenarios for staffing, resource allocation, and service planning.
Professor Iosif Mporas, who leads the project, says the model is designed to estimate what will occur if no action is taken and quantify how changing regional demographics affect NHS resources. That “do nothing” scenario is especially useful. It gives leaders a baseline against which a proposed intervention can be assessed.
For example, rising chronic illness may influence admissions, community care, workforce needs, and bed occupancy over different time scales. A unified forecast can reveal how pressure moves through the system rather than treating each service as an isolated queue.
The result is decision support, not an automatic capacity plan. Leaders still determine which actions are affordable, clinically appropriate, and operationally achievable. The model makes consequences more visible; it does not choose the values behind a decision.
Wider Data Can Improve and Complicate the Model
The Hertfordshire and West Essex Integrated Care Board serves 1.6 million residents and is preparing to merge with two neighbouring boards to form the Central East Integrated Care Board. The next development phase will incorporate data from this larger population.
More information can improve forecasts by representing a wider range of demand and demographic change. It also raises the work required to align definitions, update schedules, access permissions, and data quality across organisations. Two hospitals may record similar events differently, while missing workforce data can distort an apparently precise projection.
Demographic variables require particular care. Age, deprivation, ethnicity, and gender can reveal unequal health needs, but they can also reproduce historical inequity if patterns are treated as fixed destiny. Teams should test performance across populations and distinguish prediction from explanation.
Forecast accuracy will also change when policies, disease patterns, or service availability change. Monitoring must identify drift, and planners need to know how old the latest reliable data is.
Operational Value Depends on Decisions
A forecasting system creates no efficiency by itself. Value arrives only when leaders can connect an alert to workforce planning, commissioning, bed allocation, community support, or another concrete action.
Useful measures should therefore go beyond statistical accuracy. The NHS partners can track whether forecasts arrive early enough to influence decisions, whether resource plans improve, how often managers override the model, and whether outcomes differ across patient groups. A technically accurate projection that cannot change a budget or staffing schedule may have limited operational value.
Expansion into care homes and community services is important because hospital pressure often reflects conditions elsewhere. A system-wide view can help avoid optimising one setting by shifting the burden to another.
The Hertfordshire project presents AI as a planning instrument rather than a clinical oracle. Its ambition is deliberately practical: use existing evidence to see pressure sooner and evaluate the cost of waiting.
If it succeeds, the most meaningful result will not be a dramatic prediction. It will be a less dramatic winter, a better-timed workforce decision, or capacity created before patients experience the shortage. In healthcare operations, prevention often looks like nothing happened—and forecasting may help explain why.
