The fish feed business represents the large majority of Skretting’s sales. However, it is a business that is a highly seasonal one, dominated by a handful of major clients with a lot of bargaining power. Managing production during the peak demand season has been a significant challenge for Skretting North America because of several factors, too.
From historical data to improved forecast methodology
Analyzing a variety of historical data including sales, manufacturing, inventory and margins, allowed the team to engineer several relevant features that would form the basis of the MTS selection process. After clustering and classifying these items and identifying product groups with favorable MTS characteristics, the recommendation engine was approached as an optimization problem, suggesting an optimal production volume given the existing production and capacity constraints.
As for the improved forecast methodology, a number of statistical and machine learning models were considered and trained on the available historical sales. More specifically, different variations of the Seasonal AutoRegressive Integrated Moving Average (SARIMA) model, which attempts to predict future sales based on their history, were taken into account by employing a champion approach and selecting the best-performing model on an Item-by-Item basis.
The MTS Pro Tool
The final solution presented to Skretting consists of an interactive MTS Pro application, allowing the firm’s production planners to visualize and inspect the recommendations coming out of the MTS engine as well as offering the possibility to explore the improved underlying forecast.
The tool’s MTS Recommendation panel prompts the user to fill in the current month’s constraints such as storage capacity as well as individual production capacity per line, identifying the optimal item selection before providing the production planner with an overview of the selection’s characteristics as well as their associated recommended volumes.
The Forecast view showcases the improved forecast for the upcoming 12 months, as well as displaying both Skretting’s and Adaptfy’s forecasts against the actual sales during that period.