The volume of mobile data traffic is exploding and 5G will have to fulfil a growing variety of performance requirements, ranging from extreme mobile broadband to low-latency automotive IoT. To accommodate such demands, enhanced flexibility in managing the infrastructure is needed. Network slicing allows operators to customise resources on a per-service basis, by virtually partitioning the physical infrastructure, thereby enabling new lucrative revenue streams. However, without deep intelligence into the traffic flowing over slices, and where it originates, it is impossible to effectively and efficiently monetize them. To address this need, we are developing Microscope, an AI tool for mobile traffic decomposition. Microscope is a lightweight method that identifies and quantifies the nature and source of individual streams (e.g. Netflix, Google cloud services, Facebook, etc; down to individual base station) from aggregate streams. This allows data to be collected in the cloud rather than from expensive location-based probes, and also works with encrypted data sources which are impossible to analyse with traditional approaches such as using Deep Packet Inspection (DPI). Our technology tackles the challenges of decomposition through deep learning, due to its effectiveness in operating on large-scale mobile traffic in real-time, as demonstrated by our own research. Microscope is fast, cheap, encryption agnostic, scalable, and compatible with current NFV and open RAN initiatives