The lack of adequate measures to capture relevant factors, and the prevalence of measurement error in existing ones, often constitute the main impediment to robust policy evaluation. Random assignment of a given treatment, when feasible, may allow for the identification of causal effects, given that the necessary measurements are available. Measurement challenges include: (a) adequately measuring outcomes of interest; (b) measuring factors that relate to the mechanisms of estimated impacts; and (c) conducting a robust evaluation in areas where the RCT methodology is not feasible. In this paper, we discuss three categories of related approaches to innovation in the use of data and measurements relevant for evaluation: the creation of new measures, the use of multiple measures, and the use of machine learning algorithms. We motivate the relevance of each of the categories by providing a series of detailed examples of cases where each approach has proved useful in impact evaluations. We discuss the challenges and risks involved in each strategy and conclude with an outline of promising directions for future work.