Abstract: Crowdshipping has gained attention as an emerging delivery model thanks to advantages such as flexibility and an asset-light structure. Yet, it chronically suffers from a lackof mechanisms to create and exploit consolidation opportunities, limiting its efficiency and scalability. This work contributes to the literature in two ways: first, by introducing a novel consolidation concept that augments crowdshipping networks with dedicated scheduled delivery lines; and second, by developing a new methodology for solving complex network design problems, exemplified by the case studied in this paper, where design decisions needs to be made while taking complex operational dynamics into account, yet realistic representations of these dynamics are difficult to incorporate compactly into mathematical models. To address this challenge, we embed a neural network surrogate within a mathematical programming framework. The surrogate is based on a novel architecture that explicitly exploits the binary nature of design decisions, a feature common to many such problems. We establish theoretical guarantees regarding approximation accuracy and proximity to the true optimum. Numerical experiments with real-world data demonstrate both the practical benefits of the proposed hybrid delivery model—reducing costs, traffic, and emissions—and the computational advantages of the proposed solution framework.