In this process, you create the forecast profiles on which automatic calculation of the forecast is based.
Forecast profiles contain the information that the system requires to carry out a forecast based on a forecast model.
You have considered the following points:
· How many products do I wish to forecast?
· If I want to forecast several products, do I wish to forecast each product individually, or are there certain product families that I can forecast together?
You can assign a master forecast profile to a selection. You can thus forecast, for instance, at product group level. The results can then be disaggregated to the individual products. See Selections in Forecasting.
· Do I have historical data for the products I wish to forecast?
You can import historical data to an InfoCube. For more details, see DP Data Mart.
· What is the quality of this historical data? Does it contain errors? Is data missing? For how many periods do I have historical data?
The quality of a forecast depends primarily on the quality of the historical data. If the historical data is erroneous or is subject to factors that are not relevant to the forecast, the forecast is subject to inaccuracies.
· Is the present market situation similar enough to the past situation that I can use existing historical data? Or has there been some major structural or economic change (for example, German reunification) that renders my historical data unsuitable for forecasting future demand?
· If I have no historical data for a product, is it possible to forecast that product using the historical data of another product?
· For which products do I already know the pattern of historical data, that is, am I aware of trends or seasonal patterns in the data?
· Do I wish the system to pick the best forecasting method? For which products?
· Is demand for any of my products intermittent? Which ones?
For intermittent demand, you can use the Croston Method for forecasting.
· Do I have outliers in my historical data? Are these due to promotions or to other factors?
· Do I wish to adjust the different number of days in different months?
· Am I aware of multiple causal factors that explain the demand for a product?
In this case, you can use causal analysis, often referred to as multiple linear regression.
· Do different demand planners use different forecasting methods?
· Do I wish to combine several forecasting methods? For which products?
In this case, you use composite forecasting.
· Do I wish to model the start-up phase or the end-of-life phase of one or more products?
· Does my company intend to run promotions for any products?
See also Background Reading.
1. Decide which products to forecast and which models to use.
Deciding which model to use is another critical step. Using a model that does not fit the historical data is a major source of errors.
Determine the best model in interactive planning for a small amount of representative data. You have two options for doing this:
¡ Manually, by forecasting using various models and recording the errors produced by each model. You can then select the model that results in the smallest errors. For more information, see Forecast Comparison.
¡ Automatically, by selecting one of the two automatic model selection procedures.
2. Create a master forecast profile.
3. Create a univariate profile and/or an MLR profile and/or a composite profile.
4. If necessary, maintain the assignment of forecast profiles to selection IDs (selection variants) by choosing Goto ® Assignment in the master forecast profile. For example, you might have a selection ID for each product family you forecast and different forecast profiles for different product families.
5. Create alert profiles in the Alert Monitor that specify which alerts can be viewed by which users.