Overview of Feature-Based Forecasting for New Products

This topic provides an overview of feature-based forecasting for new products.

In feature-based forecasting, you generate a forecast for a new product by using a machine-learning algorithm that learns from the product, location, and time varying features and historical demand of existing, similar products.

The machine-learning algorithm is in the Oracle Internet of Things Intelligent Applications Cloud Service and calculates the shape and volume of launch demand for the new product. The algorithm considers how the demand for existing products varies by features across sales locations and applies these insights while generating the forecast for the new product.

The recommendations, which consist of the estimated accuracy and key feature groups for the forecast, are provided through the Planning Advisor in demand plans that use feature-based forecasting. A link is also provided in the recommendations to a page layout in which you can make adjustments to the generated forecast.

You can generate recommendations for the new product using multiple feature-based forecasting profiles.

Feature-based forecasting provides you with these benefits:

• You can automate the process of generating the forecast for the introduction of the new product.

  In contrast, while using the functionality for managing product launches, you must identify similar products and use all or a portion of their historical demand to come up with the forecast for the new product. This approach is manual and time-consuming.

• You get an insight into the feature groups of existing products that contributed the most to the generated forecast for the new product.

Note: Feature-based forecasting can be used in only demand or demand and supply plans. In the rest of this section on feature-based forecasting, "demand plan" also means "demand and supply plan."

When you use feature-based forecasting, you can extract data and generate a forecast at only the lowest level of the Product, Organization, Customer, and Demand Class dimensions and the planning time level of the demand plan. The forecasting time level set on the Demand tab of the Plan Options page for the demand plan isn't applicable to feature-based forecasting.

Prerequisites for Feature-Based Forecasting

These are the prerequisites for feature-based forecasting:

• You must have licenses for the IoT Intelligent Applications Cloud Service and Oracle Cloud Infrastructure Object Storage service.

• You must have a rich set of data for existing products and their features so that the machine-learning algorithm can learn better about how each feature is contributing to the demand shape and volume.

  Otherwise, a meaningful forecast won't be generated for the new product.

• The planning time level of demand plans that use feature-based forecasting can't be set to a period. The supported planning time levels are day, week, and month.

Architecture for Feature-Based Forecasting

In this section, the architecture for feature-based forecasting is discussed.

The data for feature-based forecasting is extracted from Oracle Demand Management through the Oracle Business Intelligence Cloud Connector and sent to the OCI Object Storage service. The data files are then routed to the IoT Intelligent Applications Cloud Service, where the machine-learning algorithm is triggered off through REST APIs by Demand Management.