Forecast · 2020
Demand sensing
Can a weekly store forecast beat the spreadsheet merchants already trust?
Method
Baseline forecast for staple lines, merchant override for promotions, and an error report the merchants could see before the Monday meeting. No hidden model.
Finding
The model won on staples and lost on promotions. Shipping it as a hybrid was the result, not a compromise we apologized for.
What moved
Forecast error on staple lines fell 18%. Promotional lines stayed human on purpose.