Forecasting demand with historical data

Knowing how much will be needed next week or next month helps with stock, staffing and cash planning. Historical data is the starting point.

Understand the patterns

Plot the history. Look for trend, seasonality (weekly, monthly, annual), holidays and one-off events.

Start simple

Moving averages and seasonal methods are easy to understand and often surprisingly accurate. Use them as the baseline.

Add complexity carefully

Machine learning models can include more factors — promotions, prices, weather — but need more data and monitoring. Use them only if they beat the baseline.

Measure accuracy honestly

Test forecasts on periods the model did not see and track forecast error over time.

Combine with judgement

Planners know about upcoming events the data does not. Let them adjust forecasts, and record why.

Forecast at the right level

Forecast where decisions are made — by product family and location — rather than at a level too detailed to be reliable.

See our predictive analytics services.