Sales Forecasting by Aggregation of Items

When a supply chain professional goes and asks a sales person about the expected sales amount for the upcoming months, the conversation usually follows like this:

-Salesman: “Sales will be rocketing, probably around $5M next month … “.

OK, but a supply chain manager does not distribute the banknotes or dollars, rather he supplies the products.

So what is the product level forecast?

-Salesman: “Huh, yes… $1M of this type and the rest other type…”

Very clear and to the point! Come on, we have 5000 SKU’s on the line and you tell me 2 types which is on the top product hierarchy.

For sure, the forecasts would be much more precise when they are done on aggregate items rather than for each item. So theoretically salesman is doing the right thing by aggregating the estimation numbers. Then someone should take the lead to create a grouping strategy for the items within the product hierarchy or in another way. Let’s assume the lead is taken by the Supply Chain Manager. So how can the Supply Chain Manager group the items for better sales forecasting?

There might be several ways to do it; keeping in mind that the most suitable way is grouping in a way that ungrouping is the easiest.

1) Hierarchical grouping:
A sample product hierarchy can be something like below;


It is definitely easier to estimate the sales amounts on the top product groupings. However still some calculations should be done to split the top level forecasts to down levels.

Algorithms can help by verifying each month the best estimation level for the item and additionally split from the best fitting level to down levels can also be utilized.

2) Grouping based on production process:

Very simple but most possibly unrealistic demonstration is shown below as an example. One type of output from some processes is going through a latest step of production, color pigments are added and then the differentiated products occur.


For that kind of a process, definitely the N type of items as a final output should not be estimated, however it will be easier to make the forecast on the level as an output of welding process and color pigments can be kept outside regular planning process and managed by safety stock strategy.

Similar to this case, there might be plenty of processes that differentiate countable number of items entering to uncountable output. Of course in the real production process, it will not be this easy to understand in which case the experience of production teams will definitely help.

3) Grouping based on raw materials usage:

This strategy works well for the companies coping with long lead times for raw materials. Just like the case explained in the 2nd point above, the supply chain manager can group the items using the same raw materials under one group and try to make the forecasts on the group instead if each item.

It will ease the life for the purchasing process and still the supply chain manager and the sales team will have enough time to decide on the final production items while the raw material is on transit.

4) Grouping based on replaceable items:

When the sales team has the ability to push the customers to replace the orders in between similar items, then a supply chain manager can use this as an advantage to forecast those replaceable items as one. Here the crucial point might be, the market should not realize that as a strategy and used for every order!

All of the methodologies listed above would help to minimize the forecast errors, however all types of aggregation will need a disaggregation at the end. For sure, the aggregation methods can be increased depending on the processes and the company structure.

A.Selim Kayacan