It is well known in data science that ‘correlation is not causation’. However, standard data science and machine learning methods are about correlation, not causation. Therefore, to answer questions about cause and effect (e.g. about measuring impacts, uplift, or about why a KPI moved), which are central to many data science projects across sectors, a new, causal, approach is needed, of which the standard data science methods are just one part. In this talk, I will show how we are using the new science of graphical causal inference at dunnhumby to answer cause and effect questions in retail, and I will share key tips and learnings.