Techniques and systems are provided for predictive modeling based on interactions with a network device. For example, a method may include receiving interaction data corresponding to a set of interactions associated with a function performed by a network device. Contextual data corresponding to a set of contexts can be received from a different source. The interaction and contextual data can be analyzed to identify a correlation between a context and an interaction. A prediction can be generated using the identified correlation. Interaction data corresponding to a new set of interactions and contextual data corresponding to a new set of contexts including the correlated context can be received. The prediction can be tested to determine whether the network device performed the function when the correlated context occurred. Contextual data corresponding to a present set of contexts including the correlated context can be received, and an automation request message can be transmitted.