As discussed in the last BFD post, sometimes a difficult and expensive problem doesn't have to be solved -- it can simply be avoided.
When the teetering boulder threatened the baby below the cliff, we removed the baby and no longer had to worry about propping up the boulder. When it seems a retailer has to forecast every item, every week, at every store -- maybe they don't. Instead of generating what can be millions of item / store / week forecasts, which probably aren't going to be very accurate anyway, it may be sufficient to forecast at the item / DC level and use inventory management practices to stock the stores.
The idea is to not solve problems you don't have to solve, to not rely on forecasts when you don't have to.
Generating millions of forecasts is not a problem with current technology such as SAS Forecast Server, or SAS Forecast Server + SAS Grid Manager for extremely large-scale problems. (See this story about generating millions of forecasts at Wyndham Exchange & Rentals.) But demand patterns at highly granular levels like item / store / week tend to be very erratic and not necessarily "forecastable" to the degree of accuracy desired.
At a SAS Global Forum presentation last month, Curry Hilton of the rapidly growing hobby farm retailer Tractor Supply Company reported that half their store / item combinations had 11 or fewer non-zero sales weeks per year. (So those store / item combinations sold zero on 41 or more weeks of the year!) Another retailer I'm familiar with has stated that half of their items sell less than one unit per week per store.
Generating store / item forecasts for some items can make sense -- when volumes are high enough and fairly regular, or there are big promotional spikes upcoming, or the items are of particularly high value. But often we don't have to solve that difficult forecasting problem, and can instead rely on something simple like a two-bin inventory control system to drive store inventory replenishment.
We'll investigate the two-bin system in Part 2, using an everyday example.
