Clustering is a form of unsupervised learning designed to group a set of data based on its inherent structural patterns and characteristics, meant to be found through the exploration of the data rather than imposed by assumptions that may introduce bias. Based on this principle, we implement a clustering methodology on functional data to identify different energy consumption behaviors of Colombian companies. We prepare the data with several preprocessing steps, focusing on preserving the original dimensions and predominant characteristics of the dataset. Subsequently, we extend multivariate clustering methods to functional data from two families: density-based (HDBSCAN) and distance-based (PAM and $k$-means). The best method is selected based on run time and internal validation measures. We subject the best method to a stability test by varying the sample size to observe consistent cluster results. Finally, we evaluate the behavior of the best method on new unseen data. Our findings show that density-based methods outperform distance-based methods with HDBSCAN selected as the best method. Three distinct types of energy consumption are identified (day, night, and constant). HDBSCAN was consistent with lower sample sizes and behaved as expected on unseen data.