Helioseismic holography is a useful method to detect active regions on the Sun's far side and improve space weather forecasts. We aim to improve helioseismic holography by using a clear formulation of the problem, an accurate forward solver in the frequency domain, and a better understanding of the noise properties. Building on the work of Lindsey et al., we define the forward- and backward-propagated wave fields (ingression and egression) in terms of a Green's function. This Green's function is computed using an accurate forward solver in the frequency domain. We analyse overlapping segments of 31 hr of SDO/HMI dopplergrams, with a cadence of 24 hr. Phase shifts between the ingression and the egression are measured and averaged to detect active regions on the far side. The phase maps are compared with direct EUV intensity maps from STEREO/EUVI. We confirm that medium-size active regions can be detected on the far side with high confidence. Their evolution (and possible emergence) can be monitored on a daily time scale. Seismic maps averaged over 3 days provide an active region detection rate as high as 75% and a false discovery rate only as low as 7%, for active regions with areas above one thousandth of an hemisphere. For a large part, these improvements can be attributed to the use of a complete Green's function (all skips) and to the use of all observations on the front side (full pupil). Improved helioseismic holography enables the study of the evolution of medium-size active regions on the Sun's far side.