Low-Power Wide-Area Networks (LPWANs) are a key technology for connecting Things to the Internet. The LoRaWAN protocol is a particularly popular example, featuring over 300 million connected devices, 5.9 million wireless receivers installed, and nearly 200 public network operators. We consider the design and operation of these networks through the lens of operations research, employing modeling tools, optimization methods, and the mindset of data-driven decision-making, to develop a toolkit for planning and operating LPWANS in a principled approach. First, we formulate learnable models for both wireless connectivity and interference. Our work on interference features a new interpretable subset choice model with strong foundation in random utility theory. Secondly, leaning on data-derived insights, we formulate a wireless receiver placement problem as a covering integer program, which can be stylized as a set cover problem. Motivated by geometric regularities in LoRaWAN connectivity, we develop a new algorithm for geometric set cover, improving the time-complexity of the state-of-the art, while matching the best known asymptotic approximation-ratio with respect to the shallow-cell complexity. Finally, we develop a new provably optimal cost-sharing mechanism for the more general covering integer program that uses duality in a strengthened LP-formulation. We use the mechanism to better understand and guide cost-, and infrastructure-sharing between LPWANs.