Videns had the opportunity to work with Tilton, a leading manufacturer specializing in the eco-design and production of plastic packaging for the food and pharmaceutical industries, to implement best practices for sales forecasting using advanced AI techniques.
Forecasting future orders and sales is a complex, often manual process that needs to be performed regularly.
Demand forecasting, also known as “Demand Forecasting,” addresses several scenarios, such as:
According to David Paquet, Industrial Management Controller, "The support solution proposed by Videns reduced the risk associated with integrating a new demand planning system, trained internal resources on time series, and achieved substantial savings in implementation costs compared to a custom solution."
The project aimed to increase the reliability and efficiency of the production plan through precise demand forecasting and to ensure the internal planning team is upskilled in using the new tools implemented.
One of the main challenges is determining the optimal prediction frequency, as it can be annual, monthly, or even weekly.
"We initially thought that the optimal prediction basis would be weekly, but Videns' expertise in time series allowed us to reassess this assumption, which proved to be suboptimal in our business context," added David Paquet, Industrial Management Controller.
It is crucial to mention that a more granular prediction does not necessarily equate to greater accuracy. At Tilton, obtaining monthly predictions leads to more informed decision-making. A model’s performance is dictated by the quality and history (or volume) of available data.
Tilton achieved substantial savings in implementation costs for its new demand planning system, with a deployment time of less than 6 months.
Videns’ expertise in time series analysis optimized the forecasting frequency and improved prediction accuracy, leading to more informed decisions.
A better understanding of substitute products and the temporal relationships between supplier purchases and sales was attained.
The chosen solution utilizes proven artificial intelligence models based on classical statistical approaches. These methods achieve the desired performance within a few weeks. In some cases, combining classical statistical methods, machine learning, and deep learning allows for modelling series with more complex signals. This multi-faceted approach also considers interactions between time series, providing a more precise understanding of substitute products and the temporal relationship between supplier purchases and sales.
Videns’ expertise in time series offers several advantages for optimizing demand forecasting, including:
Finding the right balance between the granularity of demand forecasting and forecast performance is central to the applied expertise in time series modelling.
The application of AI to demand forecasting presents significant potential for improving accuracy and efficiency in planning processes.
Additional benefits of applying AI to demand forecasting include:
