E-commerce delivery deadlines are getting increasingly tight, driven by a growing ‘I-want-it-now’ instant gratification mindset of consumers and the desire of online and omnichannel retailers to capitalize on the growth of on-demand e-commerce. On-demand deliveries with delivery deadlines as tight as one or two hours force companies to rethink their last-mile distribution network, since tight delivery deadlines require decentralization of order picking and inventory holding to ensure close proximity to consumers. This fundamentally changes the strategic design process of last-mile distribution networks. We study the impact of incorporating inventory order-up-to level decisions into the strategic design process of last-mile distribution networks with tight delivery deadlines. We develop an approximate inventory model by including an estimate of the cost of late delivery and additional transportation due to local stock-outs in a newsvendor formulation. Such local stock-outs require an order to be delivered from a more distant facility, which may lead to late delivery and additional transportation cost. We integrate our approximate inventory model and a location–allocation mixed-integer program that determines optimal facility locations, associated order-up-to inventory levels, and fleet composition, into a metamodel simulation-based optimization approach. Our numerical analyses demonstrate that pooling the additional online inventory with brick-and-mortar (B&M) inventories leads to cannibalization by the B&M network and higher B&M service levels. However, the pooling benefits to the online network outweigh the cost of inventory cannibalization. Furthermore, we show under which circumstances omnichannel retailers may have an incentive to consolidate online inventory in specific B&M facilities.