We analyze the optimal mix of ex ante experimentation and ex post learning for
the dynamic adoption of activities with uncertain payoffs in a two-phase model of information diffusion. In a first preintroduction phase, costly experimentation is undertaken to decide whether to adopt an activity or abandon experimentation. In a second stage following adoption, learning can continue possibly at a different pace while the activity remains in place; the withdrawal option is exercised following the accumulation of sufficiently bad news. We compare from a law and economics perspective the performance of three regulatory frameworks commonly adopted to govern private experimentation and adoption incentives: liability, withdrawal, and authorization regulation. Liability should be preempted to avoid chilling of activities that generate large positive externalities consistent with the preemption doctrine. Liability should be used to discourage excessive experimentation for activities that generate small positive externalities. Authorization regulation should be lenient whenever it is used consistent with the organization of regulation in a number of areas, ranging from product safety to antitrust.