Firms often face choices about when to upgrade and what to upgrade to. We discuss this in the context of upgrading to a new technology (for example, a new computer system), but it applies equally to the upgrading of processes (for example, a new organizational structure) or to individual choices (for example, buying a new car). This paper uses an experimental approach to determine how people address such problems, with a particular focus on the impact of information flows. Specifically, subjects face a multi]round decision, choosing when (if ever) to upgrade from the status quo to either a safe or a risky new technology. The safe technology yields more than the status quo, and the risky technology may yield either less than the status quo or more than the safe technology. Every round, subjects who have not yet upgraded receive noisy information about the true quality of the risky technology. Our focus on the timing of endogenous choice is novel and differentiates the results from previous experimental papers on herding and cascades. We find that, in the single]person decision problem, subjects tend to wait too long before choosing (relative to optimal behavior). In the second treatment, they observe payoff]irrelevant choices of other subjects. This turns out to induce slightly faster decisions, so the girrationalityh of fads actually improves profits in our framework. In the third and final treatment, subjects observe payoff]relevant choices of other subjects (that is, others who have the same value for the risky technology but independent private signals). Behavior here is very similar to the second treatment, so having grealh information does not seem to have a strong marginal effect. Overall we find that social learning, whether or not the behavior of others is truly informative, plays a large role in upgrade decisions and hence in technology diffusion.