Models of social evolution and the evolution of helping have been classified in numerous ways. Two categorical differences have, however, escaped attention in the field. Models tend not to justify why they use a particular assumption structure about who helps whom: a large number of authors model peer-to-peer cooperation of essentially identical individuals, probably for reasons of mathematical convenience; others are inspired by particular cooperatively breeding species, and tend to assume unidirectional help where subordinates help a dominant breed more efficiently. Choices regarding what the help achieves (i.e. which life-history trait of the helped individual is improved) are similarly made without much comment: fecundity benefits are much more commonly modelled than survival enhancements, despite evidence that these may interact when the helped individual can perform life-history reallocations (load-lightening and related phenomena). We review our current theoretical understanding of effects revealed when explicitly asking ‘who helps whom to achieve what’, from models of mutual aid in partnerships to the very few models that explicitly contrast the strength of selection to help enhance another individual's fecundity or survival. As a result of idiosyncratic modelling choices in contemporary literature, including the varying degree to which demographic consequences are made explicit, there is surprisingly little agreement on what types of help are predicted to evolve most easily. We outline promising future directions to fill this gap.