Estruturação na coleta de dados e proposição de método combinado para previsão de demanda em uma indústria de bebidas.

Abstract

Demand forecasting is fundamental for organizations and it is characterized as the beginning of strategic planning. Success in forecasting demand is the result of accurate data collection and the collection of reliable evidence about the future market trend. When well designed, it can be a determining factor in a company's competitiveness vis-à-vis its competitors. The present work sought to minimize the variability in demand forecast in a beverage industry located in southern Brazil. The study was carried out to forecast demand for the mineral water product. The methodology was exploratory and descriptive through literature review and the integration of multivariate analysis with the opinion of experts. Currently, forecast variations reach 31% compared to real demand, with the Absolute Average Percentage of Errors (MAPE) being around 15% with a standard deviation of 2%. Seasonality is one of the main factors that impact the sector, in addition to causal variables that hinder predictability. The error in forecasting the demand causes the entire production chain to suffer financial and social impacts affecting the organization as a whole. In this context, the objective of this work was to develop a combined method, called Multivariate Analysis and Inference from Human Experts (MAIHE). The results obtained by this model proved to be more efficient than those achieved by the Autoregressive Integrated Moving Average (ARIMA) model, which the organization currently uses. The MAIHE method showed an error of 14% while ARIMA reached an error of 31%, which is also a superior result compared to most studies on the topic found in the literature.