OPTIMIZING SUPPORT CENTER STAFFING THROUGH FORECASTING: A CASE STUDY COMPARING PROPHET AND LSTM

This thesis studies the performance of different forecasting models for the number of email and phone cases arriving at Arla Foods' customer support center in Sweden, Denmark, and Finland. These forecasts are to support decision making when staffing these support centers as efficient staffing affects customer satisfaction but can also cost a lot of money. The aim is to evaluate what model performs best for short-term (daily) and long-term (weekly) forecasting. The models compared are the Prophet, Long-Short-Term-Memory (LSTM) and simple Naïve models. The evaluation of these models is based on a walk-forward validation for a robust evaluation. A Diebold-Mariano (DM) test is used to evaluate whether there is a statistically significant difference in forecast performance between the models. The results show that the Prophet model consistently performs better than the other models in daily forecasts and significantly better for all datasets except one. For the weekly forecasts, the LSTM provided the best results, but these are less convincing, probably due to lack of data.