Persistent poductivity differences (PPD) exist between firms that operate in similar environments and with similar technologies. A fast growing literature is trying to explain these PPD with many different and complementary methods: by matching survey results to productivity data, by exploring the internal economis of firms, and by running randomized controlled experiments. I will provide an overview of these different approaches, and provide some ideas about how to work together with firms in order to better understand why some firms are more efficient than others. I put specific emphasis on the benefits of having both long-run information before and after experimentation.