Many populations exhibit boom-bust dynamics in which abundance fluctuates dramatically over time. Past research has focused on identifying whether the cause of fluctuations is primarily exogenous, e.g., environmental stochasticity coupled with weak density dependence, or endogenous, e.g., over-compensatory density dependence. Far fewer studies have addressed whether the mechanism responsible for boom-bust dynamics matters with respect to at-risk species management. Here, we ask whether the best strategy for restoring habitat across a landscape differs under exogenously vs. endogenously driven boom-bust dynamics. We used spatially explicit individual-based models to assess how butterfly populations governed by the two mechanisms would respond to habitat restoration strategies that varied in the level of resource patchiness, from a single large patch to multiple patches spaced at different distances. Our models showed that the restoration strategy that minimized extinction risk and boom-bust dynamics would be markedly different depending on the governing mechanism. Exogenously governed populations fared best in a single large habitat patch, whereas for endogenously driven populations, boom-bust dynamics were dampened and extinction risk declined when the total restored area was split into multiple patches with low to moderate inter-patch spacing. Adding environmental stochasticity to the endogenous model did not alter this result. Habitat fragmentation lowered extinction risk in the endogenously driven populations by reducing their growth rate, precluding both "boom" phases and, more importantly, "bust" phases. Our findings suggest that (1) successful restoration will depend on understanding the causes of fluctuations in at-risk populations, (2) the level and pattern of spatiotemporal environmental heterogeneity will also affect the ideal management approach, and (3) counterintuitively, for at-risk species with endogenously governed boom-bust dynamics, lowering the intrinsic population growth rate may decrease extinction risk.