Forecasting for Consumer Products

I used to work for a cosmetics company in a competitive local market. The meeting that I dreaded the most every month was the sales forecast meeting: a chaotic collision between sales, marketing and finance control. The method we used was,
1. Use historical data for forecast the upcoming sales for the SKU
2. Adjust the forecast by promotional activities
3. For new items, forecast was based on the number of stores the new item will pipe in and historical data on items in the same subcategory.
4. The forecast then was reviewed by the controller to make sure that it is in-line with the annual sales goal.

There were constant disputes on how the adjustment based on promotional activities are made. As a marketer, I had to go in the meeting to show all the evidence to justify a forecast number that I am not sure of.

This paper (Yes!A paper on consumer product forecasting in the proceedings of World Congress in Engineering and Computer Science!) talks about a mixed forecasting model for consumer products in the Thai market. The model considers a mix of SKU, promotional mechanisms, and media.
1. Forecast for existing SKU is based on historical sales, adjusted by promotion and media coefficients.
2. Forecast for new items is based on historical sales data in the subcategory, adjusted by promotion and media coefficients.
3. A systematic manual reviewing flag is developed to ruled out any illogical forecast results.

The model proposed by the paper was reported to significantly increase the forecasting accuracy in the Thai market. While the model proposed does not seem to be a novel one, it would be interesting to see more case studies on how mining the past data and gaining more insights about the activities result in a better forecast.