How to plan capacity and staffing with AI

Normalize demand and capacity units

Choose one operational unit, such as cases completed per week, analyst hours per review, or appointments per session. Convert demand and capacity to the same time interval and service boundary. Separate new demand, backlog, repeat work, and failure demand.

NHS England’s demand-and-capacity guidance distinguishes required capacity from available capacity and explicitly considers variability, rebooking, and ongoing remodelling in its health-service context. Its general-practice guide recommends consistent measurement over time, understanding variation, smoothing where possible, and then matching capacity to demand. These principles transfer as planning discipline, not staffing advice for another sector.

Build an input ledger:

| Input | Base | Range | Evidence | Caveat | |---|---:|---:|---|---| | New reviews per week | 80 | 65–110 | Twelve-week intake | Launch week excluded | | Productive hours per reviewer | 24 | 20–27 | Rosters and time data | Leave varies | | Hours per completed review | 2.0 | 1.6–2.8 | Sample of 42 | Complex cases cluster | | Rework share | 10% | 5–20% | Issue log | Definition recently changed | | Service target | 90% in five days | Fixed | Approved service rule | Priority cases separate |

The numbers are synthetic. Ranges preserve the observed uncertainty that an average alone would erase.

Build flow and workload scenarios

Calculate demand hours by case type, including expected rework. Calculate available capacity from scheduled productive time after planned leave, meetings, support duties, and system downtime. Do not use paid hours as fully productive hours.

Create at least base, high-demand, and constrained-capacity scenarios. A base week of 80 reviews at two hours plus ten percent rework requires about 176 review hours. Eight reviewers with 24 productive hours provide 192 hours. That apparent 16-hour margin may disappear if complex cases arrive together or one reviewer is absent.

Use AI to generate scenario tables and check arithmetic. Reconcile every conversion manually, especially percentages, weeks versus months, and people versus full-time equivalents.

Model variability and queues

Service performance depends on when work arrives and how it varies, not only total weekly volume. Plot daily or weekly demand and capacity. Identify seasonality, batch arrivals, priority work, skill constraints, and tasks that cannot be pooled.

A failure example divides average demand by average productivity and reports 7.3 people, rounded to eight. It ignores absence, peak timing, rework, and the fact that only three reviewers can handle one case type. Repair the model with ranges, skill pools, queue conditions, and explicit service consequences.

Distinguish core capacity from flexible capacity. Temporary effort can clear a backlog, but treating it as permanent may conceal an unstable operating model. Record activation lead time and limits.

Compare staffing and operating choices

Options may include changing intake rules, smoothing demand, reducing avoidable rework, cross-training, adjusting service tiers, scheduling flexible capacity, or adding permanent roles. Show the effect, cost, lead time, constraint, and residual service risk of each option.

Do not let the model prescribe employment action. Managers decide staffing with financial, workforce, legal, service, and wellbeing considerations beyond the calculation.

Verify the capacity plan

Before approval, confirm:

The artifact is the capacity model plus input ledger, demand profile, skill matrix, scenario table, option comparison, and refresh log.

• demand and capacity use the same unit and boundary;
• backlog, rework, and new work are separate;
• productive time excludes known non-delivery work;
• variability and skill constraints are represented;
• scenarios include peaks and constrained capacity;
• service consequences are visible;
• arithmetic reconciles from source inputs;
• options show lead time and residual risk; and
• the plan has a refresh trigger and owner.

Sources and claim map

The workload, service rule, and staffing values are synthetic. This method does not establish a required headcount, employment action, service result, or cost outcome.

• NHS England, Demand and capacity models: supports required-versus-available capacity, variability, and iterative remodelling in its scope.
• NHS England, How to align capacity with demand in general practice: supports consistent measurement, variation analysis, smoothing, and matching capacity to demand.

Sources: [1] [2]