Forecast Modeling capabilities in SAP APO vs. other statistical tools
I will take SAS software and Forecastpro as benchmark tools here to compare the capabilities of APO DP.
Between SAS and APO DP, the difference is huge. You have an array of statistical models to choose from in SAS while APO DP, has just a few limited set of models. Even among the models available in APO, the optimization and convergence of model selection to result in the parameters are some what inefficient and does not compare to even inexpensive off-the-shelf products like Forecast Pro or Autobox.
For univariate forecasting, you can use Exponential smoothing models, intervention models, Box-Jenkins models, distributed lag models, vector autoregression models etc. SAS allows you to model using a variety of techniques and allows you to customize them, although you require deep knowledge in statistics and programming. It is not a plug and play tool.
APO DP offers the basic exponential smoothing models and linear regression model that use a deterministic time index as the independent variable. The exponential smoothing models work for most purposes but they are not the most efficient - for example the iterations of the smoothing parameters namely alpha beta and gamma are in increments of .5 not anything in between.
And some configurations I have seen and versions I have seen can be quite sticky and may just resort to one model with all parameters set to .3 or some user-defined selection. This is also quite surprising because it assumes the user already knows what the parameter is - from where? By doing the modeling in an external statistical tool and sticking them back in?! But this is not APO's fault - mostly the configuration team that did not know how to leverage the power of the tool.
SAS also can be used for generating transfer function models when you need to do true causal models. The only option available in APO is to use the MLR models using a multi-variate profile and then inserting your events in a linear model. This may actually produce incorrect results if there is auto-correlation in your data, which is the most likely probability since most sales data will be serially auto-correlated.
So in summary, SAP has some usable statistical models but one has to be careful in choosing what to use where. You need a demand planning expert to set the profile settings and the model selection process and train your users on what to do and what not to do.
For additional information on APO DP modeling, please see
http://demandplanning.net/sapAPO_workshop.htm
Leave a Reply