Product Sales Forecasting.

Hi,

We are trying to forcast product sales for next three months based on their sales for previous 12 months. In this case, Microsoft Time Series algorithm requires the sales data to be present for each product for past 12 months (?). However, our products have typical life span of 6 months and obviously the new products will not have sales before they were added. Any help will be very much appreciated.

Thanks

Riju

Riju -

A few pointers that have been very successful for us in forecasting sales and inventory:

1. Can you build store clustering models to increase the data set size?
2. For new products you can also try correlating sales from similar existing products. We have utilized attributes from existing products and mapped their sales as a predictor for new product sales.

Here is a whitepaper that provides a detailed inventory forecasting example from Project REAL http://apollodatatech.com/company/shell.html?projreal

Hope this helps,

Jeff

Please visit Apollo Data Technologies http://apollodatatech.com/|||

Hi Jeff,

Thanks for the response.

It seems that the Out-of-stock predictive model in REAL project is based in the Decision Tree algorithm which does not require the data to be present for each period. I am having problem to use Time Series algorithm becuase my products typically have sales for about six month and i am trying to use the last 12 months data to predict the sales of next 3 months.

Thanks

Riju

|||

This must be a common problem in forcasting. Isn't there any way to handle this?

Thanks

Riju

|||You should read the paper referenced by apollo. In these circumstances you likely need to build a tree model to predict based on similar products, not use time series as there's not enough historical data.|||

Hi Jamie,

Thanks for the response.

I read the paper referenced by Apollow which gave the detail information on how to predict the product sales using the Decision Tree algorithm. However, i was trying to get some help on how to use the Time Series algorithm in my situation. Thanks to your response which stated that the Time Series algorithm is not applicable in my situation.

Riju