The sales forecasting process has been structured over time with new technologies and
tools, for data consolidation and handling. The companies, which previously had no
focus on the sales forecasting process, were not impacted by the lack of it, but currently,
adjustments are necessary for its insertion, because there is consensus that only
intuitivity, usually directed by past experiences or subjectivities, or optimized results or
underestimated them. Walking in the above, this research aims to identify a sales
forecast model appropriate to the portfolio of a medium-sized beverage company. In the
study of this dissertation, the explanatory research technique was applied with
exploratory and descriptive analyses, and minitab® and Excel software was also used®
to perform the analyses through statistical abstracts, tables and figures, so that there was
the assertive choice of the model to be applied to the business. Qualitative and
quantitative forecastmodels, graphic analysis, residue scans and forecast error
calculations were evaluated. The mean deviations and MAPEs (Mean Absolute Percent
Error) of the models were compared: moving average, exponential smoothing, linear
trend and holt winter and, as conclusion, the models with the lowest prediction errors
were: moving average N=2 with MAPE=14.8%, exponential smoothing with
MAPE=15.2% and linear trend with MAPE=15.4%. The choice was for the exponential
smoothing model, although not the slightest error is easy to apply and weights the
historical data.
|
