Agriculture & Future Food Systems · Article

Forecasts and variable food demand

A community kitchen expects a familiar number of meals, but attendance and menu choices vary. Use an explicit forecast and compare it with what actually happened.

Preparation and related reading

Shared Foundations

This article helps you

Distinguish a forecast from a known future quantity.
Describe variation without treating a short record as a guaranteed range.
Use observed errors to investigate the demand and recording basis.

Begin with a specific service

Imagine a community kitchen planning one vegetable meal each day. Its ingredient order depends partly on expected attendance, but people may arrive in different numbers or choose another meal. A useful forecast states the particular service and period rather than treating all purchases as one demand total.

Demand concerns the food or service requested or needed at the chosen boundary. Recorded sales or portions served can differ from that demand when stock runs out or a service is inaccessible. Begin with what the record actually measures before deciding which number belongs in a forecast.

Retain the basis of the record

Identify the product, food state, unit and timing. A dry-lentil purchase, cooked-meal portion and person attending are related quantities, but they are not interchangeable. A conversion requires a stated preparation or service assumption.

Note exceptional events and missing records. A festival day, changed opening hours or cancelled delivery can alter the observed pattern. Measurement error is also different from real demand variation. Extra decimal places in a forecast do not remove either kind of uncertainty.

Read a small average

Suppose three invented comparable days record 80,100 and 120 meals requested. Their mean is 100 meals, with the smallest and largest observation 40 meals apart. That average summarises this supplied record; it does not establish the next day’s demand.

The diagram compares the 100-meal forecast with several possible observed outcomes. The prior 80–120 range is not a confidence interval or a guaranteed future bound. A later 130-meal request could lie outside it. A short record can reveal variation while leaving seasonal and longer-term patterns unresolved.

One forecast can differ from observed requests

Original hypothetical comparisons, not probabilities, a prediction interval, prescribed preparation quantity or measured forecast performance.

Read the diagram as text

A hypothetical 100-meal forecast is compared with 80 and 130 requests. These outcomes show an excess or shortage relative to that number, not a probability distribution.

Parts

Forecast: 100 meals: Specific service and future period identified
Observed: 80 requests: 20 below forecast; later food fate unspecified
Observed: 130 requests: 30 above forecast; service response unassessed

Connections

Forecast: 100 meals → Observed: 80 requests: compare outcome
Forecast: 100 meals → Observed: 130 requests: compare outcome

Consider seasonal and contextual changes

Attendance, holidays, seasonal availability, price and menu suitability can influence a food-service pattern. These possibilities need relevant evidence in the actual setting. Do not assign a cause to a high-demand day from its total alone.

A forecast may use historical records alongside current information about the planned service. Longer records can reveal recurring patterns, but a past season is not automatically a model for a changed population or arrangement. Supply capability also matters: forecasting demand does not ensure that suitable food can be obtained or prepared.

Compare the forecast with the outcome

If an invented forecast is 100 meals and 130 are requested, the difference is 30 meals. If only 100 are served because no more are available, using served meals as the demand observation would hide the unmet requests. Keep the observation basis attached to the error.

Under a separate 100-meal forecast with 90 requests, the difference is 10 fewer than forecast. An excess preparation or order can follow, but its final fate requires its own stock and safe-use record. Forecast error is not automatically the quantity wasted, and the two situations can have different consequences for people and the kitchen.

Revise without erasing the earlier forecast

Save the forecast before seeing the outcome, then record what changed and whether the original basis was appropriate. Rewriting the forecast to match the observation removes useful evidence about its performance.

Use repeated comparisons to decide what needs attention: the service record, context information, prediction approach or supply response. Procurement and storage decisions must also account for lead time, capacity and uncertainty. The aim is suitable reliable meals, not a prediction score divorced from the people being served.

Worked examples

Summarise the supplied record

Three hypothetical comparable days have 80,100 and 120 meal requests. What mean and observed spread follow?

Mean = (80+100+120)/3 = 100 meals.
Observed spread = 120−80 = 40 meals.
These describe three observations, not a guaranteed future demand interval.

Mean 100 and observed spread 40 meals; no forecast accuracy or confidence level established.

Separate requested and served

An invented forecast is 100 meals;130 are requested but only 100 are served because supply ends. What demand error and record risk follow?

Requested minus forecast = 130−100 = 30 meals.
Served minus forecast = 100−100 = 0, but this hides 30 unserved requests.
Retain which count measures the demand question.

Thirty more requests than forecast; serving exactly the forecast does not prove accurate demand prediction.

Check your understanding

Must tomorrow’s requests lie within the 80–120 range from three earlier days?
Show answer and reasoning
No.
The observed range is not a guaranteed bound or confidence interval.
Does a higher request count establish that one menu change caused it?
Show answer and reasoning
No.
Attendance and other contextual differences can contribute.
Is a ten-meal forecast excess automatically ten wasted meals?
Show answer and reasoning
No.
Preparation, stock and subsequent suitable destinations need their own records.

Key ideas

State what demand and the observation count mean.
Keep forecast, actual result and context separate.
Use errors to investigate the system rather than conceal uncertainty.

References

Sources supporting the explanations and examples in this article.

WFP Logistics Cluster2 sources
1. Logistics Operational Guide: managing inventory levelsWorld Food Programme; Global Logistics ClusterReference guidelog.logcluster.orgRead source
2. Logistics Operational Guide: defining when to orderWorld Food Programme; Global Logistics ClusterReference guidelog.logcluster.orgRead source