Poverty alleviation programmes generally rely on data about local economic livelihood for identifying places with highest need for aid; nevertheless, this information traditionally comes from patchy and expensive household surveys. In this framework, the “Earth Observation for Poverty” (EO4Poverty) project has implemented a novel and robust system based on advanced deep-learning to generate accurate spatial wealth maps by exploiting freely available EO data in combination with sparse in situ survey data. In particular, this improves existing approaches and provides an easily transferable service for creating maps which can be employed as ready-to-use tool for policy makers.