Data use is becoming a prominent strategy for educational innovation and improvement across countries. However, the fragmentation of data collection often hinders the capacity of policymakers, researchers and practitioners to access and analyse the wealth of data routinely generated in educational institutions. A critical step towards realising the potential of education data is thus to set up a strong data infrastructure at the national/state level. Longitudinal data systems represent a promising solution to this challenge. We discuss the capabilities and limitations of current education data systems, drawing on a survey of 64 systems in 30 countries. We argue that the next generation of education data systems should integrate longitudinal, individual-level administrative records with learning management platforms, and incorporate an extended repertoire of analysis and reporting tools in order to support richer types of diagnosis and provide enhanced feedback to stakeholders. The potential of longitudinal data systems to foster innovation and improvement in education is illustrated by a discussion of how these tools can support educational research.