The energy sector is going through a very turbulent period since 2021. The energy crisis, which started before the war in Ukraine but was magnified by it, led to highly volatile energy prices that coincided with the post-pandemic economic recovery and the expansion of economic growth, resulting in a rather unstable situation. Still, European countries reacted with a surge in awareness of energy efficiency and in adopting energy-saving measures whilst trying to accelerate the propagation of renewable energy sources. The latter can only be successful if it is based on a cohesive energy strategy. An appropriate energy strategy has to be based on managing the demand, which in turn leads to the necessity to predict consumption values. Nowadays, this can be done with various models, which, however, must be trained. One of the tools for successfully creating energy consumption forecasts is supervised machine learning (ML). In this research, 7 different ML models were examined and trained, with the help of Python and with an input of 5 y of data or, alternatively, with real hourly energy consumption data of a tertiary sector’s building. This analysis aids the selection of the appropriate model considering internal and external factors, with a special focus on cyclicality, seasonality, and trend in both hourly and daily predictions. The results indicate that a deep learning architecture based on an artificial recurrent neural network (RNN) is chosen to better deal with hourly predictions of energy consumption. The effectiveness of the models applied to the time series data decreased in the evaluation of the daily data. However, Gradient Boosting frameworks based on decision trees show that valuable predictions are feasible. Lastly, it was proven that the ambient temperature improves the effectiveness of forecasts both in hourly and daily models.