This session will outline some of the NLP tasks and approaches that were used to build and deploy an AI system that detects sensitive information in enterprise scale. In predictive language solutions - from categorizing entire documents to extracting sensitive sentences or entities - the varying context of each sample is crucial. Yet, the encoding of context in ways that machines can interpret is still possible using sufficient labeled data and deep neural networks. The road to building useful, useable AI products goes through much more grounded technologies. Research has shown that 28% of worker's time in the office is dedicated to email work.