Demand Forecasting is science revolving around the ability to predict quantities in advance. Whether it is short-term forecast or longer into the future, it is crucial to know what to expect from customers and the market in advance and with the maximum possible accuracy.
Leveraging all available data is obviously part of the solution; however it is only part of the story. This is why our Demand Forecaster takes a comprehensive approach to the problem, understanding how the forecasting issue fits into the process and how human intervention can be just as important.
Basically the solution revolves around three major components:
- A powerful Big Data Ingestion layer to leverage internal and external data,
- A state-of-the-art forecasting modeling techniques,
- A process-oriented collaborative workbench to mix and match the analytical forecast with multi-level business savvy in order to create the best forecast possible.
The industry-standard model families are included and run against one another to determine the best model for each series or family of series, and backtesting techniques can help to fully automate the process, by selecting the best performer.
Moreover, we constantly thrive to improve our models or extend their breadth / areas of applicability.
In fact, it is possible to significantly improve on existing forecasts even without proper time series for the SKUs we elect to predict (though not without data!).
Among others, such is the case of the fashion industry, where retailers and buyers maintain a certain degree of attitude and taste in deciding their purchases, even though up to all SKUs may be different from the previous years’. Past data and reliance on such behavior is key to getting good estimates.
Also, you can leverage the ability of continuously re-estimating models, even with numbers of series (whether they may be forecasts, products, etc.) soaring into tens of thousands and more.
Users can interact with the engine based on their skill level – openness is a feature of the solution, although stability, automation and reliability remain the cornerstones of the engine.
Following the pure forecasting phase, users can interact with the results by leveraging a multi-user solution for collaboration in which each stakeholder can view and refine the forecast, based on his information and/or business sense.
The ability to run a workflow with different levels of approval as well as reconciliation of time-series / SKU hierarchies is also a fundamental aspect of the solution.