A characteristic feature of adaptive systems is their ability to be context-aware which means being able to sense changes and adjust behavior accordingly. In developing such systems, engineers would often assume at design time several situation types that are of high occurrence probability and specify again at design time corresponding system behavior types. What about other situations? In most cases, any other situation would be addressed by a (standardized) fallback behavior. Still, this goes away from context awareness because such a behavior is not aligned to the “current” situation. As noted in previous works, a mitigation in this regard would have been the provision of real-time algorithmic support and this is nevertheless insufficiently explored. Why? Because often there are too many possible-to-occur situations (besides the ones considered at design time) that constructing an algorithm would become challenging. In this regard we see solution directions in the perspective of Data Analytics, inspired by the developments in recent years and referring to previous work addressing Bayesian Modeling. The idea is that: (i) Under a fallback behavior we classify and/or cluster the “current” situation such that we establish where it “belongs” to; (ii) We then statistically derive a match between the established “situation space” and a corresponding “desired behavior space”. That is how the fallback behavior would be much more specific to the situation at hand, this leading to a more effective context-aware system. We present those ideas in the current position paper, planning more elaboration and then validations as future work.