When A/B/n tests (aka megastudies) have covariates that describe individuals, they can potentially be used to design targeting policies, i.e., to achieve better outcomes by assigning optimal interventions for specific individuals rather than applying the same intervention to everyone. We applied five popular targeting approaches to two large-scale field experiments with many interventions aimed at increasing vaccination rates (the Walmart study of [Milkman et al., 2022]) and the Penn-Geisinger study of [Milkman et al., 2021] and [Patel et al., 2023]), and found limited value of targeting in the Walmart experiment and a four times higher value of targeting in the Penn-Geisinger experiment. Our research explains the difference in the gains from personalization between the two studies and shows that neither heterogeneity in responses nor independence across potential outcomes within individuals is sufficient to predict whether a targeting exercise will be successful. Instead, a specific form of heterogeneity, which we call "actionable" heterogeneity, determines the value of targeting. Namely, to be "actionable"' for targeting, heterogeneity that moderates the treatment effects within different subgroups is not sufficient. It also needs to have a few subgroups for which the most successful intervention is not the same. Visually, this appears as a crossover between one intervention and another, if the individuals are ranked by their treatment effects. We show that the magnitude of crossovers with the best-performing intervention determines what portion of heterogeneity is "actionable"' for targeting.