Predicting quantities ahead.
WHAT TO ESTABLISH
- The horizon: how far ahead
- The granularity: per product, per location, per period
- The purpose: purchasing, staffing, cash planning
WHY GRANULARITY MATTERS
Forecasting total sales is far easier than forecasting each item, and the purpose determines which is needed.
WHAT DRIVES ACCURACY
Seasonality captured correctly Known future events included Sufficient history
WHAT KNOWN FUTURE EVENTS ARE
Holidays, promotions, price changes, closures.
WHY THEY MATTER SO MUCH
They explain variation the model cannot otherwise learn.
WHAT TO ALWAYS COMPARE AGAINST
The naive forecast, and whatever the business currently does.
WHY THAT SECOND ONE
An experienced buyer's judgement is frequently a strong baseline.
WHAT INTERMITTENT DEMAND IS
Many periods with zero sales.
WHY IT IS DIFFICULT
Standard methods handle it badly, and the average is misleading.
WHAT TO MEASURE
Error in units that matter commercially: stockouts and excess.
WHY NOT PERCENTAGE ERROR
It behaves badly near zero, which is exactly where intermittent demand sits.
WHAT TO PRODUCE
A range, not a single number.
