ML / auto statistical cut-off / reference product assignment in IBP / Manage Product Lifecycle?

[vincentverheyen]

I understand the SAP IBP Manage Product Lifecycle app can be used to produce automated forecasts for the future demand of new products introduced by manufacturers into their product portfolio on a regular basis, when for these new products the historic values needed for forecasting don't exist. Therefore, using the Manage Product Lifecycle app, one can search for products that are presumed to have (had) a similar historical sales pattern and assign them as reference products to the new product.

Some further correction of the historical data is necessary, as I understand that some reference products might stop being sold at a certain point in time, and this would skew the data. Therefor the manual feature of Validity Dates have been introduced in the Manage Product Lifecyle app.

This manual assignment of Validity Dates seems quite time-consuming and to require a lot of manual effort. My question is: in order to automatize the assignment of reference products to new products, what alternatives exist within IBP to save human time and effort spent on assigning both the reference products to the new products, as well as on adjusting the Validity Dates / data correction of the historical values of the reference products. I am aware of the Upload and Download feature, so third-party tools might be able to be integrated this way, and am also aware of the possibility to run the forecasts on an aggregated level to minimize data fluctuation and skewing, but I wonder if there is any other built-in tools available to meet the required automatization.

Is there any possibility for automatized statistical analysis of the reference products, for example if a certain statistical value representing an 90% or more lower average value of the Sales of the reference product during 4 consecutive periods is reached, that the reference product automatically becomes unassigned (or Validity Dates adjusted)? How to tweak the existing statistical algorithms within the Manage Product Lifecycle app?

Or are there any further elaborate statistical techniques or even Machine Learning (ML) within IBP to calculate the future demand forecasts of a new product based on data of earlier products that are presumed to share certain characteristics? Are the only ways in IBP to - out of the box - assign reference products to new products manual assignment

- via Excel (then upload to the Manage Product Lifecycle app);
- or manually within the Manage Product Lifecycle app Fiori interface directly?

Is there no smarter integrated way in IBP to, for example use a Kibana Elasticsearch style database which creates a similarity predictor index for each node/product on which to automatically base itself for a truly automated forecast for the future demand of a new product without the product having any existing historical values - which could also integrate tendency data such as external news / Google Search hits / ...?

Or how can we leverage Machine Learning (ML) capabilities for such newly introduced articles without historical data?

[LauraTozzo]

Hi Vincent,

there are a lot of questions in this post, so let me try to go through them one by one.
In the following I tried to rephrase the questions based on my understanding.

• How to automatize the assignment of reference products?
  I understand you would like to have some (ML) tools to automatically find a reference product without having to assign it manually. This is a very interesting requirement but such a capability is not available in IBP.
• How to automate the "unassignment" of a reference product from the new product?
  If I understand this correctly, you would like the phase-in end date to adjust automatically based on some rules (e.g. the forecast for the new product is >90% of the sales of the reference product for 4 consecutive periods). For this you could set up a custom alert that informs you when such an event has happened and allows you to manually adjust the phase-in end date accordingly.

• How can one assign reference products to new products in IBP?
  There are three ways: via upload, via Manage Product Lifecycle App (as you mentioned) and via OData Service https://api.sap.com/api/IBP_Product_Lifecycle_ODataService/overview.
• Is there a way in IBP to use a Kibana Elasticsearch style database which creates a similarity predictor index for each node/product on which to automatically base itself for a truly automated forecast for the future demand of a new product? Which could also integrate tendency data such as external news / Google Search hits / ...?
  I am not particularly familiar with Kibana Elasticsearch, but to be able to do this you would have to duplicate all the data of all products into the external system. You can use the existing tools to extract data from IBP and then use these data in the external system.
  Regarding tendency data, you can already now use additional data to determine the forecast using Forecasting Models such as MLR, ARIMAX and Gradient Boosting (ML). To do this you need to integrate these data as key figures in IBP.
• How can we leverage Machine Learning (ML) capabilities for such newly introduced articles without historical data?
  You can use the Gradient Boosting Algorithm to learn from the history of the reference product(s) and generate a forecast for the new product. At the moment additional data (called independent variables) are not supported for new products.
  Later this year you will have the opportunity to use any custom algorithm in an external system to generate the forecast in a tightly integrated fashion, and so you could use ML capabilities that learn from the reference product (and even based on additional inputs) and apply the learnings to generate the forecast of the new product.

Hope this helps

Best regards

Laura