Comments on: Ask Dr. Jain: How Do I Get Started With Holt Winters Exponential Smoothing?

By: Anonymous

Anonymous — Mon, 04 Jun 2018 21:44:36 +0000

Normally, if we cannot forecast well at a SKU level, we forecast it at a category level. With this, data becomes more stable, and forecast gets better. The data become stable because there are always some SKUs whose sale are going up, while of others going down. When they are aggregated, they offset each other. If there are some SKUs in a category that cannibalize others, prepare their forecasts separate from those being cannibalized. This will further improve the category level forecast. For example, fume free oven cleaner cannibalizes the sales of a regular over cleaner, yet both belong to the same category. Prepare forecast of regular oven cleaner separately.

By: Giridhar Prabhu

Giridhar Prabhu — Fri, 01 Jun 2018 06:18:33 +0000

One of our Customers had this bottom-up approach to Forecasting at SKU level. But after years doing it this way, they felt it was too much work for too little benefit and that they get similar/better results from a higher level forecast as Dr. Jain points out.

By: Thea van Zyl

Thea van Zyl — Thu, 31 May 2018 12:23:22 +0000

Dr,
What would the approach be if you have SKU’s within the category that does not behave the same as the majority of the SKU’s.