Data collection is the starting point for a useful forecasting process. Gather demand rather than sales, and collect it at the level where decisions occur: for example, weekly demand by warehouse instead of monthly demand by market when inventory is deployed weekly. section 14.4.1
Use the collection effort to improve decisions without delaying practical progress:
- Clean relevant demand data before judging a model; poor data can outweigh sophisticated modeling. section 14.4.1
- Match the data’s aggregation and horizon to inventory deployment and other operational decisions. section 14.4.1
- Add demand-driver data selectively. First test what is already available, then weigh the cost and expected insight of further collection and cleaning. section 14.4.1 section 14.2.2
- Treat external datasets cautiously when their timing, granularity, acquisition, or cleaning practices do not fit the requirement. section 14.4.1
Choosing useful demand-driver data
A manufacturer may keep important information in Excel files and must decide whether collecting and cleaning additional demand-driver data is worth the expected forecasting improvement. Weather illustrates the difficulty: ice-cream demand may rise with temperature, but the relationship can flatten or reverse at extremes, so more data does not automatically produce a simple or reliable signal. section 14.2.2
