Skip to content
← Back to glossary

Workforce Demand Forecasting

Workforce demand forecasting estimates how much work will arrive and when it will arrive. The demand unit can be calls, chats, tickets, orders, visits, cases, or tasks. The forecast should use the same time interval, queue, channel, location, and skill detail needed for staffing decisions.

Demand forecasting is not the same as staffing forecasting. Demand forecasting predicts the work. Staffing forecasting converts that demand into people or hours after average handling time, occupancy, service targets, shrinkage, skills, and uncertainty are applied.

Workforce demand forecasting process

  1. Collect consistent historical demand by the planning interval.
  2. Mark missing data, system outages, one-time anomalies, and known events.
  3. Backtest suitable methods on recent periods that were not used to fit the model.
  4. Add known future drivers, such as campaigns, holidays, launches, or opening-hour changes.
  5. Publish a baseline forecast and an uncertainty range by queue or work type.
  6. Measure WAPE for error size and bias for repeated over- or under-forecasting.
  7. Convert the approved demand forecast into staffing requirements and schedule coverage.

Common workforce demand forecasting methods

Useful methods include comparable-period averages, seasonal naive forecasts, exponential smoothing, ARIMA models, daily profiles, and multi-seasonal methods. The best method depends on the operation's data. Test models on the operation's own history instead of selecting one only because it is more complex.

Why this matters for planners and team leads

A weekly or monthly total can hide the peaks that create queues, overtime, and missed work. Forecast at the level where staffing can change. Contact centers may need 15-minute or 30-minute intervals, while project or field teams may make decisions by day or week.

Review demand error separately from staffing error. A demand forecast can be accurate while staffing still fails because handling time, shrinkage, attendance, skills, or the schedule were wrong. Keeping the layers separate makes the corrective action clear.

Example in practice

A support operation normally receives 80 calls between 09:00 and 09:30 on Monday. A recent product release creates a repeatable 25% increase, so the adjusted demand forecast is 100 calls for that interval. The planner keeps the release adjustment separate from the baseline so its effect can be reviewed later.

After the interval ends, actual demand is 108 calls. The team records the forecast error and checks whether the miss was isolated or part of a repeated low bias. The approved demand forecast then feeds handling time, queueing, shrinkage, and skill calculations for the staffing plan.

Frequently asked questions

What is workforce demand forecasting?
It estimates how much work will arrive and when. The demand unit can be calls, chats, tickets, orders, visits, cases or tasks, and the forecast should carry the same interval, queue, channel, location and skill detail the staffing decision needs.
How is it different from staffing forecasting?
Demand forecasting predicts the work. Staffing forecasting converts that demand into people or hours after average handling time, occupancy, service targets, shrinkage, skills and uncertainty are applied. Confusing the two produces a headcount that was never derived from a service target.
What does the process look like?
Collect consistent history by planning interval, mark missing data, outages, anomalies and known events, backtest candidate methods on recent periods not used to fit them, add known future drivers such as campaigns or opening-hour changes, then publish a baseline with an uncertainty range by queue or work type.
Which accuracy measures should be used?
WAPE for the size of the error and bias for repeated over- or under-forecasting. They answer different questions: a forecast can have acceptable error while being consistently low, which staffs the operation short every period.
Why backtest on held-out periods?
Because a method judged on the data used to fit it will always look good. Comparing methods on periods they did not see is the only way to know which one will behave on next month, which is the only month that matters.
Should anomalies be removed from history?
Mark them, do not delete them. Outages and one-off events should be excluded from the baseline, but real peaks the operation may see again belong in the history. Cleaning out every spike produces a forecast that has never seen a busy day.

Put this into practice

See how Soon handles workforce demand forecasting in your shift scheduling workflow.

See the product
How to forecast staffing needsWorkforce forecasting software