Content area
Full Text
Sales forecasting is important for a company to plan its production. The quality of its forecasts influences finances and the product availability. The impact of sales forecasts on a company may result on an immobilization of cash flow by causing a high stock level, which is the opposite of out-of-stock impact. The purpose of this study was to find a suitable model for predicting the best company sales forecasts that has a better accuracy or production plan. The proposed method includes an adjustment of the prediction model by including the key account managers' expertise as qualitative forecasting method. This adjustment was analyzed using different time series forecasting techniques such as exponential smoothing, seasonal autoregressive integrated moving average and Facebook Prophet. These techniques were compared in parallel with neural network approaches such as long-short term memory. Comparisons were made using root mean square error and residual stock to determine whether the forecasts were too optimistic or pessimistic. The proposed model is dynamic. Adjustments of the qualitative inputs could directly influence the proposed values obtained using different quantitative methods.
Keywords: demand forecast, exponential smoothing, SARIMA (seasonal autoregressive integrated moving average), Facebook Prophet, LSTM (long-short term memory), KAM (key account manager)
(ProQuest: ... denotes formulae omitted.)
INTRODUCTION
Sales forecasting is becoming an important subject even in small and medium enterprises (SMEs). Poor predictions have several negative impacts for companies, such as overstock (immobilization of cash flow) or stockout and lack of components. Improving sales accuracy means also improving the company's future business projections (Haberleitner et al., 2010).
For decades, it has been proved many times that quantitative forecasting models provide better results than qualitative forecasting models (Ramosaj & Widmer, 2020). The most well-known quantitative model is time series forecasting, in which historical observations are collected and analysed to develop an applicable model. The goal of time series forecasting has often been to improve forecast accuracy (Siami- Namini et al., 2018). Time series methods have been applied to improve different areas such as forecasting the power load for the electricity market (Bozkurt et al., 2017), wind energy production (Hui et al., 2012), food retail demand (Pereira Da Veiga et al., 2014), road and traffic optimisation (Zhao et al., 2017), cryptocurrency exchange rates (Chen et al., 2021), COVID-19 infection...