Academic journal article The Journal of Business Forecasting

Forecasting after an Organizational Realignment

Academic journal article The Journal of Business Forecasting

Forecasting after an Organizational Realignment

Article excerpt

Organizational realignments result from changes in product groupings, customer / sales geography, or distribution networks ... realignments alter the hierarchical structure of the historical data used for statistical forecasting ... hierarchical changes can be problematic for forecasting software packages ... storing historical demand at the lowest level makes it possible to derive the organizational groupings appropriate for forecasting.

The statistical approach to forecasting can be very efficient: Historical demand data are fed into the statistical forecasting software, demand patterns are modeled, and forecasts are generated. With good software, this entire process can be automated. If the demand happens to be reasonably "well behaved" (without too much erratic behavior and randomness), then you can even get good forecasts.

A key assumption in statistical forecasting is that the demand being modeled is appropriate to the future you are trying to forecast. In other words, when you need to forecast demand for bananas, it is best to model the actual historical demand for bananas, rather than the historical demand for oranges, peaches, hotdogs, automobiles, orpersonal computers.

A common problem in business forecasting is that the appropriate demand history does not exist for what you are trying to forecast. A special case of this is when forecasting for an entirely new type of product, or when expanding your sales or distribution into an entirely new territory where you have no history at all to go on. This article is not going to address the special case of new products or geographies. Instead, it will address the problem of demand history that has been "contaminated" by organizational realignment. When this happens, the history is no longer appropriate to use in your statistical forecasting models. Some examples will help to clarify the problem.


Organizational changes occur all the time-at least on an annual basis for most companies. Sometimes these changes are on the product side, where individual items are reorganized into different groups or when new levels of groupings are added. Table 1 shows organizational realignment that occurred when Items 003 and 004 were moved to a new Group C. Other organizational realignments result from changes in the customer, sales geography, or distribution network because of:

* Reorganizing the sales force into new regions and territories, hire/fire sales reps, etc.

* Closing of a Distribution Center (DC) and assigning of its customers to the remaining DCs for order fill.

* Opening of a new DC and shifting of customers from existing DCs to a new one for order fulfillment, an example of which is shown in Figure 1. Here some customers of DC1 and DC2 were reassigned to a newly opened DC3.

If statistical forecasting is done at the most granular level of detail, such as by individual item at individual customer, then realignments wouldn't be a problem. Realignments don't change history at the most granular level of detail. However, for many good reasons, we don't always build our forecasting models at the most granular level. Statistical models are often built at some intermediate or high level of aggregation, such as at a product group (rather than an individual item) or at a sales region (rather than individual customer). This is because behavior that is intermittent and seemingly random at the most granular level will often exhibit a forecastable pattern when aggregated to a higher level. For good forecasting, it is often best to model the pattern at some intermediate or higher level of aggregation, and then allocate it back down to the most granular level. (This is known as middleout forecasting when you start at an intermediate level of aggregation, or topdown forecasting when you model at the highest level of aggregation.)

While realignments don't change history at the most granular level, they do change the history for the aggregated levels. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.