Pollinator-mediated selection on plants can favor transitions to a new pollinator depending on the relative abundances and efficiencies of pollinators present in the community. A frequently observed example is the transition from bee pollination to hummingbird pollination. We present a population genetic model that examines whether the ability to inbreed can influence evolutionary change in traits that underlie pollinator attraction. We find that a transition to a more efficient, but less abundant pollinator is favored under a broadened set of ecological conditions if plants are capable of delayed selfing rather than obligately outcrossing. Delayed selfing allows plants carrying an allele that attracts the novel pollinator to reproduce even when this pollinator is rare, providing reproductive assurance. In addition, delayed selfing weakens the effects of Haldane's sieve by increasing the fixation probability for recessive alleles that confer adaptation to the new pollinator. Our model provides novel insight into the paradoxical abundance of recessive mutations in adaptation to hummingbird attraction. It further predicts that transitions to efficient but less abundant pollinators (such as hummingbirds in certain communities) should disproportionately occur in self-compatible lineages. Currently available mating system datasets are consistent with this prediction and we suggest future areas of research that will enable a rigorous test of this theory.