Nowadays, the Web offers huge amounts of data sources for the benefit of the community. However, there is a lack of practical approach for converting and linking multi-origin data sources into one coherent smart data set. In this paper, we define a service-oriented architecture to attach explicit semantics to data, to solve heterogeneity issues, and to remove data inconsistencies in order to convert raw documents to quality Linked Data. We motivate the need for a service oriented architecture for smart data with a live scenario based on the Audience Labs company information system. We show how our service-oriented architecture adapts to the company needs and facilitates semantic annotation, data integration and exploitation of the resulting smart data.