New innovations, practices, and behaviors often spread through diffusion processes in which earlier adopters influence later adopters. However, research on diffusion has a well-documented success bias — cases in which a new innovation successfully spreads through a population or organizational field garner more attention and theorizing than the countless other cases in which similar innovations fail to take off. The same theories and models that account for successful diffusion often become cumbersome when tasked with explaining failed diffusion, an outcome that is at least equally common. In this talk, I will present results from a theoretically-informed computational model of organizational behavior to argue that failed diffusion need not be more mysterious than successful diffusion. In fact, both outcomes may reflect the same underlying mechanisms rooted in how actors form social expectations for how others will behave. Organizations interpret their peers’ decisions to adopt or reject a new innovation in light of their own socially formed expectations, with unsurprising decisions having less impact than conspicuous surprises. Consequently, successive adoptions of a new innovation reinforce its spread while also paradoxically making its continued diffusion more susceptible to disruptions that can make a previously growing “bandwagon” suddenly and unexpectedly collapse. These dynamics make the spread of new innovations noisy and unpredictable because the same innovation facing identical initial conditions can diffuse widely in some cases but fail to launch in others. While we often think of institutionalized expectations as making the social world more predictable, the opposite also holds—widely-believed prophecies can be self-defeating as well as self-fulfilling.