Forecasting a new SKU
When no history is available
In recent posts we’ve talked about basic time series methods and advanced time series methods for sales forecasting, we’ve highlighted the core differences between demand and sales forecasting. But in all of these blog pages we’ve assumed that enough history of an SKU is available in order to get a good forecast and, obviously, it is not always the case. How can we deal with a very very short history or with a brand new product in the inventory?
Things get hard when there is no sales data, especially for a tool who relies on it, as it is for a demand forecasting software. The more history, the more accuracy of the predictions. A thumb rule which is usually true.
However, the time series of sales is not the only source of value. But we’ve to analyze different cases.
This is the case of a new model of a product series. Let’s think about a smartphone: the new model is appearing in the market with huge investments in marketing by the manufacturers. The first thing that comes to mind is that the new SKU will replace the previous and the missing time series history could be inherited from the old SKU. Oh, awesome, close to victory.
However, the new smartphone will cost more than the previous in the first weeks, thus it’s not plausible that the sales of the old model will drop to zero instantly. That is, the movement of sales from the old SKU to the new SKU is gradual and it is commonly called cannibalization. Thus, a demand forecasting software can exploit such relationship and do a good forecast, although several factors have to be considered and it deserves a distinct blog post.
A new SKU comes with a lot of information
Even if no history is available, since it is new, an SKU comes with a lot of useful information. Here we’re going to briefly introduce a few ideas on how to get data to work with.
A new SKU, let’s say a new model of white ripped jeans, has got intrinsic data. Firstly, it is a white jeans, then it is a jeans. Oh, will it be sold in Arizona? Yes, let’s have a look at the sales history of white jeans in Arizona. Uh, a lot of information. And this information can be combined and clustered in order to get insights on our brand new white ripped jeans future demand.
Finally, no matter if no sales history is available. In the era of Big Data, the main concern is to extract value from data, whichever is the source or the semantic. For a new SKU, several solutions can be exploited in order to get a good forecast.