Abstract In this paper I share some key principles and examples from the Data Informed Practice Improvement Project. The project develops an innovative model of mathematics teacher development, with three main strands: teachers work with data from their classrooms, they use this data to understand and engage with learner errors in mathematics, and they do this collectively in professional learning communities, with facilitation from members of the project team. I describe each of these strands and present three examples that show how the strands work together to support teacher learning in professional communities. These examples illuminate a key achievement of the project: a focus on learner errors in professional learning communities can develop powerful conceptual knowledge of mathematics among teachers at the same time as developing teachers’ knowledge for teaching and teaching practices.