This document discusses the need for systems science approaches to scale up improvements to livelihood systems in smallholder agriculture. It argues that agronomic recommendations do not account for the complexity of livelihood contexts. A systems approach considers interactions between crops, livestock, trees, households and landscapes. Options must be tested across a range of biophysical and socioeconomic conditions to develop a nuanced understanding of what works where. Participatory methods, monitoring of a wide range of options and aligning with development projects and policy are key to scaling up effectively while accounting for local variation.