Neyman’s seminal paper in 1923, which introduced the potential outcome framework and the analysis of randomized experiments, has arguably laid the foundation of causal inference for cross-sectional data. For time-series data, the framework of randomization inference is far less well-understood due to the interference: the potential outcomes at a particular time typically depend on treatments assigned before that time. Motivated by the literature of N-of-1 trials in clinical research and sequential AB testing in online marketing, in this talk, we study randomization experiments and causal inference when N = 1, borrowing insights from system identification and probability theory. The talk is based on joint work with Benjamin Recht (UC Berkeley).