5th Generation (5G) radio access technology aims at supporting services with different latency requirements in the same air interface. The usage of an early Hybrid Automatic Repeat reQuest (HARQ) feedback based on a prediction of the decoder's outcome allows reducing the over-the-air latency, with particular advantages for mission critical applications. However, predicting the ACK/NACK feedback before detection occurs is unfortunately error prone. In this paper, we discuss the potential of using an early HARQ feedback in 5G networks considering the occurrence of wrong predictions. Two options are studied: only the early feedback is transmitted, and both early and regular feedback are transmitted, with the possibility of the latter one of correcting wrong early predictions. Numerical analysis shows the potential of the early feedback of reducing the over-the-air latency up to 50%. Wrong estimates can, however, significantly affect outage rate and latency for the case of only early feedback, while their impact is very minor in case the regular feedback is empowered to correct them.