FAQ: Retail Store Forecasting

Self-anointed as The Aristocrat of forecasting related blogs, The BFD takes its public service role seriously (or at least as seriously as anyone who self-anoints on a regular basis can be taken). Among the services The BFD provides, both publicly (on webcasts or speaking engagements) and privat… ely (by phone, email, or personal consultation in all manner of exotic and seedy locales), is to answer questions on the mysteries of business forecasting. One of the commonly heard questions is about retail store forecasting, as in the exchange below:

Q: We are a retailer trying to forecast store/item demand every week, and we aren’t doing very well. How do we get better forecasts at this level of granularity?

A: Demand is often intermittent (lots of zeroes) and erratic at the most granular level of detail. You may be unable to get a good statistical model at that level, and even if you can find an appropriate model, you may not be able to expect highly accurate forecasts.

It is often better to focus your forecasting efforts at some intermediate level, and then apportion the forecast down to the most granular levels. Better still, you may be able to forecast and manage inventory, capacity, and so forth at that intermediate level, and use supply chain management techniques to handle the levels below.

Consider an example from retail. You may have thousands of items and hundreds (or thousands) of stores. One option is to try to forecast each of these store/item combinations each week, and you can do this with large-scale automated forecasting software, but may still not get highly accurate forecasts. Instead, consider the real business objective, which is to have the appropriate amount of inventory on store shelves to meet customer demand, without excess inventory and without lost sales. Often the cause of store-level stockouts is poor replenishment practices from the store’s warehouses. If the warehouses aren’t managing inventory properly, or aren’t able to distribute it promptly, then it may not matter how well you can forecast as the store level. Shrink (loss, theft, or damage) of product at the store level can be another problem.

A reasonable approach in this situation is to focus your forecasting efforts at the item/warehouse level. You have a much better chance of having good forecasts at this intermediate level (aggregating all the stores that pull from the warehouse). As long as you maintain the appropriate level of inventory at the warehouse, and are able to promptly fill orders from the stores, then shelf inventory in the stores should be managed properly. You may be much more effective in meeting your overall inventory and service objectives by using good inventory management practices, rather than expecting to solve the store/item issues just by better forecasting at that level. (You could have perfect forecasts at the store/item level, but if your warehouse can’t fulfill replenishment orders in a timely fashion, the perfect store/item forecasts are of little value.)

Goldratt has applied his theory of constraints approach to the retail problem in a recent book (Isn’t It Obvious?), which despite several type-setting issues, is a good and fast read. The key point is that there are often non-forecasting related solutions to business problems. In the retail store problem, the desire is to have the appropriate amount of inventory on store shelves. If we could reasonably expect to have highly accurate forecasts by store/item, that would be great, but we probably cannot hold this expectation. So instead, we find that a better way to address the problem is to forecast at a level where we may have some hope of sufficiently accurate forecasts (e.g. at the warehouse or distribution center), and then use good stocking and replenishment practices to maintain the appropriate inventory in the stores. Forecasting by item/store becomes unnecessary, or at least is no longer critical to addressing the business problem. Remember Aphorism 6 from The BFD (the book):

**** Minimize the organization’s reliance on forecasting *****