The growing threat of drug resistance has inspired a surge in evolution-based strategies for optimizing the efficacy of antibiotics. One promising approach involves harnessing collateral sensitivity-the increased susceptibility to one drug accompanying resistance to a different drug-to mitigate the spread of resistance. Unfortunately, because the mechanisms of collateral sensitivity are diverse and often poorly understood, the systematic design of multi-drug treatments based on these evolutionary trade-offs is extraordinarily difficult. In this work, we provide an extensive phenotypic characterization of collateral drug effects in E. faecalis, a gram-positive species among the leading causes of nosocomial infections. By combining parallel experimental evolution with high-throughput dose-response measurements, we provide quantitative profiles of collateral sensitivity and resistance for a total of 900 mutant-drug combinations. We find that collateral effects are pervasive but difficult to predict, as even mutants selected by the same drug can exhibit qualitatively different profiles of collateral sensitivity. Overall, variability in collateral profiles is strongly correlated with the final level of resistance to the selecting drug. In addition, collateral effects to certain drugs (e.g. ceftriaxone) are considerably more variable than those to other drugs (e.g. fosfomycin), even for drugs from the same class. Remarkably, however, the sensitivity profiles cluster into statistically similar groups characterized by selecting drugs with similar mechanisms. To exploit the underlying statistical structure in the collateral profiles, we develop a simple mathematical framework based on a Markov decision process (MDP) to identify optimal antibiotic cycling policies that maximize expected collateral sensitivity. Importantly, these cycles can be tuned to optimize long-term treatment outcomes, leading to drug sequences that may produce long-term collateral sensitivity at the expense of short-term collateral resistance.