A system (1) has propensity models (4), and a controller (5) for automatically supplying key event data for user interactions with a network. The key event data is distributed asynchronously around a network to the model systems (4). The model systems (4) use the key event data and user data from subscriber systems to generate subscriber analysis outputs such as propensity to take one of a set of pre-configured actions. The probed data includes usage, charging, and purchase history data. The models are weighted sub-models, in which weights are based on user segment.