Worked example
Situation
Several stores combine sales and stock in spreadsheets. The buyer spends time preparing the report, while seasonality, promotions and stock-outs blur into one sales history. It is hard to tell what to order and how reliable a forecast is.
What changes
Automatic data collection and checks, one shared report and a demand forecast with uncertainty ranges. The model is compared with a simple seasonal rule; if it is not better, the simple rule stays.
Where people stay in control
The buyer sees the exceptions and approves each order, taking lead times, minimum batches and cash limits into account.
What changes for the team
Instead of an evening building the spreadsheet, the buyer reviews exceptions; the forecast shows which assumptions deserve a second look.
How the example is calculated
- Before
- 24 h/month
- After
- 8 h/month
Preliminary estimate · month
4 cycles × 6 h = 24 h; 4 × 1.5 h + 2 h of data checks = 8 h
Assumptions
4 preparation cycles a month; 6 hours now; 1.5 hours after; 2 hours of data checks. Lower stock-outs and overstock are not claimed in advance.
What to check in a pilot
Forecast error on held-out periods against the baseline; no leakage of future data; preparation time; stock-outs and overstock under comparable conditions. Zero sales during a stock-out are not zero demand.