A commonly used approach to explore astrophysical sources below the detection threshold is image-domain stacking or co-adding. This uses known positions of a source population sample identified at one observing wavelength to make statistical measurements of the sample at a different wavelength, where the images are not sufficiently deep for direct detections of the individual objects. These samples are typically selected through intrinsic or observed properties such as stellar mass or optical colours in an attempt to limit biases, maximise completeness, or separate out sub-populations of interest. We explore the utility of an alternative approach by designing an algorithm (using a non-linear neutral controller) to select subsets of the input parent sample of galaxies based on what we refer to as “goal-orientated stacking” objectives. In this case, we set the goal as identifying a subset of galaxies with physically correlated properties (e.g. stellar mass, redshift, star formation rate) that maximise the radio continuum signal-to-noise level. We explore a few applications of this alternative approach and discuss possible extensions.