I found the paper to be quite exciting, and it is also solving one of the main tasks in anomaly detection, that is real-time anomaly detection. With other machine/deep learning based algorithms, one can model the data distribution quite efficiently, but there is a tradeoff of flagging time. With the statistical method suggested in the paper, you not only get state-of-the-art results but are also able to report the results in real-time, a big plus.