In the context of Industrial Internet of Thing (IIoT) applications, network administrators must use their available bandwidth to both identify best 5G NR (New Radio) configuration and best transmission schemes. In this paper we derive optimal channel access and packet replication schemes for Ultra Reliable Low Latency Communications (URLLC) traffic in 5G and Beyond networks. Based on typical system configurations, we identify three main regimes, characterized by the delay and the consequent presence or absence of feedback received from the network. In particular, we identify (a) extreme situations where there is only one time slot opportunity for packet transmission with possible replications in the frequency dimension, (b) blind repeated replication with several transmission slots opportunities, with no received ACK before the delay expiration, and (c) far-sighted scenarios where, additionally, some ACKs may be received, acknowledging older replicas of the packet. We propose adapted replication models for each radio scenario and develop corresponding mathematical models from which the transmission schemes may be optimized. The proposed policies are semi-distributed, in the sense that the optimal policy is communicated to each device by the network, that then applies it autonomously in order to respect the stringent timing constraints of URLLC. We then show the corresponding system dimensioning for achieving the target reliability and identify the best NR configurations to be deployed in the controlled industrial environments.