Conventionally spatial normalization is done after individual subject analysis in AFNI. One rationale is that by postponing any pre-processing steps as late as possible you have the individual subject data available in original space if you want to view/analyze/present individual subject data in original space. Another advantage is that it saves a lot of runtime running 3dDeconvolve (30-60 times less computer time cost) compared to spatial warping onto tlrc space with resolution of 1X1X1 mm3 before running 3dDeconvolve.

But this does not have to be the case, and the use can alternatively convert individual subject data into tlrc space in AFNI before 3dDeconvolve. Such a flexibility lets the user decide whatever order is more comfortable. In fact a big advantage of warping the individual data into common space is that the investigator can easily compare group results with individual activation maps. There should not have any significant difference in terms of group analysis results. The reason is the following. Spatial normalization involves warping which is not a linear process in terms of original coordinate system. However, for any specific voxel in tlrc space the warping process is indeed a linear step, and in this sense it would be commutative with other linear steps. Thus normalizing before or after individual subject analysis does not make any significant difference in terms of final group analysis results.