We consider a robust beamforming problem where large amount of downlink (DL) channel state information (CSI) data available at a multiple antenna access point (AP) is used to improve the link quality to a user equipment (UE) for beyond-5G and 6G applications such as environment-specific initial access (IA) or wireless power transfer (WPT). As the DL CSI available at the current instant may be imperfect or outdated, we propose a novel scheme which utilizes the (unknown) correlation between the antenna domain and physical domain to localize the possible future UE positions from the historical CSI database. Then, we develop a codebook design procedure to maximize the minimum sum beamforming gain to that localized CSI neighborhood. We also incorporate a UE specific parameter to enlarge the neighborhood to robustify the link further. We adopt an indoor channel model to demonstrate the performance of our solution, and benchmark against a usually optimal (but now sub-optimal due to outdated CSI) maximum ratio transmission (MRT) and a subspace based method.We numerically show that our algorithm outperforms the other methods by a large margin. This shows that customized environment-specific solutions are important to solve many future wireless applications, and we have paved the way to develop further data-driven approaches.