Design, Development, and Implementationof a Demand Forecast Model

Abstract

This thesis project studied the demand planning process of Cargill Global Edible Oil Solutions (GEOS) in the EMEA region and had the objective of developing a demand forecast model for the Retail Food Services customers to improve their satisfaction levels. The scope of this covered the commercial and contractual relationships, for the delivery of the bottle oil ready for distribution at the retail market; the procurement and production processes, as well as the Sales and Operation Planning (S&OP) process in detail. For the S&OP, Cargill relies on the Make to Forecast (MTF) production strategy, which has two main mechanisms for demand management and production planning. The first one is the use of a demand forecast and the second is the management of safety stocks. The demand forecast helps to guide the production planning to meet the upcoming weekly demand until it becomes available 7 days before the delivery time. While the safety stocks help to compensate for forecast errors and deviations, as well as for production setbacks that may occur. Unfortunately, for Cargill GEOS the automatic demand forecast has an accuracy of 42%, and the safety stock levels rose considerably and were managed on a constant level all year round to compensate. This impeded Cargill to meet the customer demand on time and in full according to the contracted service levels.
The main challenges in this situation were the limited data availability and the demand data fragmentation due to changes in the product specifications. Which required the development of specific procedures and frameworks to align the fragmented data and to obtain a weekly demand forecast. For the development of the forecast model, 3 statistical seasonal forecasting models were considered to maintain the transparency of the process, the ARIMA model, the ETS exponential smoothing, and the STL+ETS. Due to the limited data availability, the chosen model was the STL+ETS model as this was the one that was able to replicate the observed demand, with an average weighted accuracy of 76%.
The insights obtained from the seasonal forecast allowed to improve the safety stocks management, from a fixed level all year round to a seasonal management according to the expected demand and its variability throughout the year.
The solutions developed in this thesis project were not limited to the demand forecast development but they aimed to improve the demand management process in different aspects. Which will allow Cargill GEOS to improve considerably their service levels without considerable additional costs, such as the increase of the production or warehouse capacity.