Nowadays, usage-based acquisition models for e-books, as PDA and EBA, are quite spread and allow libraries to offer wider collections, without acquiring them. A PDA with Ebook Central was implemented in 2019 and a first EBA with Wiley in 2021. The EPFL Library ran several analyses in order to evaluate their accuracy after one or more years, with respect to users’ needs. The advantages and limitations of the use of usage statistics in collection development were also highlighted in this study. These two models have proven to be a good complement to the acquisition models based on liaison librarians’ e-book selections. Both also greatly alleviate this selection work, while increasing the management and analysis tasks for the librarians in charge of electronic resources. The PDA model has required adjustments over the years, to ensure that accessible collections contents better match to EPFL domains. The usage of the titles acquired through this model is generally higher than that of those acquired otherwise, nevertheless the PDA presents an increased risk of acquiring titles with an ephemeral popularity, despite the implementation of a loan system before the final purchase. The EBA program showed an interest among patrons in titles related topics and publication years that were not necessarily expected. This acquisition model requires a large investment for statistical analysis by the librarians responsible of electronic resources, but the results of these analysis allowed to go beyond the sole need to select titles at the end of the EBA program. These two models led us questioning more broadly the evolution of the role of selection made by librarians, and the part of human intervention in models where usage statistics are the starting point for the selection.