A fundamental aspect of well performing cities is successful public spaces. For centuries, understanding these places has been limited to sporadic observations and laborious data collection. This study proposes a novel methodology to analyze citywide, discrete urban spaces using highly accurate anonymized telecom data and machine learning algorithms. Through superposition of human dynamics and urban features, this work aims to expose clear correlations between the design of the city and the behavioral patterns of its users. Geolocated telecom data, obtained for the state of Andorra, were initially analyzed to identify “stay-points”—events in which cellular devices remain within a certain roaming distance for a given length of time. These stay-points were then further analyzed to find clusters of activity characterized in terms of their size, persistence, and diversity. Multivariate linear regression models were used to identify associations between the formation of these clusters and various urban features such as urban morphology or land-use within a 25–50 meters resolution. Some of the urban features that were found to be highly related to the creation of large, diverse and long-lasting clusters were the presence of service and entertainment amenities, natural water features, and the betweenness centrality of the road network; others, such as educational and park amenities were shown to have a negative impact. Ultimately, this study suggests a “reversed urbanism” methodology: an evidence-based approach to urban design, planning, and decision making, in which human behavioral patterns are instilled as a foundational design tool for inferring the success rates of highly performative urban places.