Feedback transmissions are used to acknowledge correct packet reception, trigger erroneous packet re-transmissions, and adapt transmission parameters (e.g., rate and power). Despite the paramount role of feedback in establishing reliable communication links, the majority of the literature overlooks its impact by assuming genie-aided systems relying on flawless and instantaneous feedback. An idealistic feedback assumption is no longer valid for large-scale Internet of Things (IoT), which has energy-constrained devices, susceptible to interference, and serves delay-sensitive applications. Furthermore, feedback-free operation is necessitated for IoT receivers with stringent energy constraints. In this context, this paper explicitly accounts for the impact of feedback in energy-constrained and delay-sensitive large-scale IoT networks. We consider a time-slotted system with closed-loop and open-loop rate adaptation schemes, where packets are fragmented to operate at a reliable transmission rate satisfying packet delivery deadlines. In the closed-loop scheme, the delivery of each fragment is acknowledged through an error-prone feedback channel. The open-loop scheme has no feedback mechanism, and hence, a predetermined fragment repetition strategy is employed to improve transmission reliability. Using tools from stochastic geometry and queueing theory, we develop a novel spatiotemporal framework to optimize the number of fragments for both schemes and repetitions for the open-loop scheme. To this end, we quantify the impact of feedback on the network performance in terms of transmission reliability, latency, and energy consumption.