SPP - Forecast - Analyse the results

[Former Member]

Hi

We are using SPP in SCM 7.0, and we are in the process of upgrading to EHP3 shortly.

We use Planning service Manager and execute the forecast.

Automatic model selection is being selected.

Our Business wants to understand in a better way the Forecast results.

In other words, they want to cross check the Forecast, if it is correct.

I am trying my best to explain them, but still the job is   half done only

Dear Experts,

Can you suggest me a way out how to:

1. analyse and confirm the forecasting results

2. how to confirm that the model selected by SPP is correct.(since sometimes  when you do the forecasting on simulation mode, it changes)

3. how to measure the forecast accuracy, any report is there?

Thanks & regards

SRS Kannan

[Former Member]

SRS,

I think you are asking these questions to the wrong people.  Only the client and his business people can define what are the standards and KPIs that are to be measured.  Once you know what these are, then you can help them to determine the best way to analyze if these standards are met.

It sounds like someone is asking if the SAP software solution you have created meets some type of accuracy standard.  The client's business users should first define what is the desired output, and then you can help them measure if the new SAP-based solution complies.  If they have a statistical algorithm that they insist must be applied to the data, then have their statistician give you the algorithm, and then have him perform the algorithm on a sample set of data using his non-SAP trusted tool.  You can have SAP perform the forecast against the same set of sample data, and then you can compare SAP results to his results.

1. analyze and confirm the forecasting results

First, someone from your client's side must define what must be analyzed and confirmed.  Can you elaborate?

2. how to confirm that the model selected by SPP is correct.(since sometimes  when you do the forecasting on simulation mode, it changes)

First, have the client define what is 'correct'.  How does he identify a model that is NOT correct?

3. how to measure the forecast accuracy, any report is there?

The client decides how to define 'accuracy'.  What does he have to say?

I hear these questions from time to time.  Remember, a statistical forecast is, in the end, still just a forecast.  It is a prediction of the future.  As such, all forecasts are ultimately wrong, since noone here in Saṃsāra is able to tell the future.  Agonizing over precision of the forecast is time poorly spent - finding that the mathematics are wrong by a few decimal places has little impact on the business results.  Your statistical forecasts will be extraordinarily good if the forecasts match the actuals 80% of the time.  When I press business users to tell me how they want me to determine if the forecast I created for them today is 'accurate', most of them have no good answers, beyond using yet another statistical algorithm to determine fitness.  The most knowledgeable business experts will realize that they need to wait until the end of the period that the forecast covered; collect the actuals, and then compare the two using an industry standard, such as Mean Average Deviation.

If your client is sophisticated enough to have a statistical expert on staff (most are not), then he will direct you how he wants to test the forecast.  Beyond that, some clients will have some rough idea how they want to determine the quality of the forecast, and you can help them to develop measurement tools to track this quality, once you know what he wants.  The key is, the standards have to come from the business, not from IT.

Best Regards,

DB49