In the situation of the client, forecasting is used when it comes to sales target setting for the upcoming year. Sales figures are forecast in order to coordinate marketing activities, customer services and more. Currently, these forecasts are generated rather based on experience than data. Some difficulties have been experienced in this procedure, due to large customer groups or lack of experience.

The present Bachelor’s thesis addresses different aspects of time series, time series forecasting and forecast result measurement. Using the sales data of the client, the question is tackled whether an acceptable time series forecast can be produced and if so, how accurately. In order to lay a foundation of knowledge and understanding for the practical implementation, a tailored literature review was conducted. Based on the insights obtained, the time series were analyzed, a forecasting method was selected, forecasts generated and finally measured.

27 customer groups have been forecast based on daily observations. The elaboration of the forecast results revealed three main findings. Firstly, it was found that the chosen Holt-Winters’ method is not capable of producing acceptable forecasts for each of the 27 time series, even though some of them have been forecast very accurately. Secondly, a tendency that the accuracy increases when aggregating the daily forecast results into months and years could be noticed. However, caution is advised when drawing conclusions from aggregated results, because a potential risk for exceptions was discovered. Thirdly, only about one-third of the 27 time series proved to be worth the effort to be forecast with Holt-Winters’ rather complex method. 17 time series achieved a better result when forecast using the simple forecast method, in this case the simple exponential smoothing method. Concluding, it can be stated that for a data-driven forecast approach, it is necessary to use more than one forecasting method. However, this paper also demonstrates that forecasting the clients’ sales data is feasible to some extent, since some customer groups could be forecast with high accuracy.