The current fiscal climate is focusing attention on the need for more efficient government. However, we have remarkably little rigorous information on which are the most cost-effective strategies for achieving common goals like delivering high quality education in deprived neighbourhoods or reducing carbon emissions. This paper argues that randomised impact evaluations can provide an effective way to generate the information needed to make government more effective. Advances in the theory and practice of running randomised evaluations mean that a wider range of questions can be answered than ever before. Elsewhere in the world, fundamental questions are being answered about how humans behave, which in turn are being used to design new policies which themselves are rigorously tested. By learning from these results, and by conducting more randomised evaluations on issues important to UK policy (both at home and abroad) it will be possible to design more effective policies and do more with less.