The increasing complexity and volume of data mandate tighter integration between analytics and visualization. In this paper, I propose a pattern of integrating of computational and visual analytics techniques, called just-in-time (JIT) interactive analytics. JIT analytics is performed in real-time on data that users are interacting with to guide visual-analytic exploration. Fundamental to JIT analytics is enriching visualizations with annotations that describe semantics of visual features, thereby suggesting to users possible insights to examine further. To accomplish this, JIT analytics needs to 1) identify insights depicted as visual patterns such as clusters, outliers, and trends in visualizations and 2) determine the semantics of such features by considering not only attributes that are being visualized but also other attributes in data. In this paper, I describe the JIT interactive analytics pattern, along with a generic implementation for any type of visualization and data, and provide a particular implementation for point-based visualization of multivariate data. I argue that the pattern provides a useful user experience by elevating the cognitive level of interaction with data from pure perception of visual representations to understanding higher level semantics of data. As such, this supports users in building faster qualitative mental models and accelerating discovery. Furthermore, facilitating insight opens new research opportunities such as visual-analytic action recommendations, improved collaboration, and accessibility.