Changing the LEMO sales forecast by machine learning

The water bottling and distribution market is very complex and has differences associated with many elements unrelated to production itself. Seasonality due to weather, civil holidays, religious celebrations such as Ramadan and holiday periods, together with other common constraints inherent to production, bottling and distribution, forces companies to continuously investigate new ways of developing daily sales analysis and prediction models adjusted to a changing and tricky market. The need to be able to work with different lead times (distribution, manufacturing, etc.) is one of many challenges that pioneering enterprises like LEMO face, forcing them to apply innovation to their analytical processes.

In an initial phase of analysis, consolidation, integration, cleaning and adjustment, we generated a preliminary dataset that enabled us to analyse complete time series by region and product for analysis. We harnessed potential trends and seasonal variables to subsequently apply multiple analysis models that would let us pinpoint the most suitable one for the analysis type required, identify predictive and non-predictive pairs and subsequently implement them for industrialisation and consumption by multiple user profiles. 

Analytical consultancy and implementation of a forecast system with a one-week lead time as the first step in the predictive model to respond to multiple business issues at the firm. Preparation of an industrialisation model and exploitation of the implemented model, with weekly lead times but able to be adapted to the various circumstances of market changes for continuous adjustment. Provide LEMO management with a data-driven decision-making tool to reinforce its experience and intuition.