Economic activities such as crowdfunding often involve sequential interactions, observational learning, and successes contingent on achieving certain thresholds of support. To analyze them, we incorporate an all-or-nothing (AoN) feature in a classical model of information cascade. Relative to standard settings, we find that an AoN target effectively delegates early supporters’ downside protection to a later “gate-keeper”, and leads to uni-directional cascades and prevents agents’ ignoring private signals and imitating preceding agents’ rejections. Consequently, information aggregation improves, and issuance becomes less under-priced, especially with a large crowd of agents, and even when agents have the options to wait. More generally, endogenous AoN targets improve the financing efficiency of costly projects and the harnessing of the wisdom of the crowd under information cascades.