This paper discusses why the extensive scientific results on predicting embedded systems temporal behavior never, or very seldom, reaches the industrialcommunity. We also point out the main issues that the scientific community should focus on in order to facilitate industrial-strength timing predictions. The core problem is that the scientific community uses too simplistic or research oriented timing models. The models stemming from academy do not fit well with the structure of real systems. Thus, extracting a timing model that is amenable for analysis may prove prohibitively difficult. And even if a model can be extracted, it may not capture real system scenarios well. Thus, results from analyzing these models do not reflect real system behavior, leading to unnecessary pessimistic timingpredictions. In recent years, response-time analysis has matured to a degree where models can express complex system behaviors and analysis results are relatively tight with respect to real system behavior. However, in order to fully reach its potential, and be accepted by industry, several improvements of the technique are needed. First, behaviors that are commonly used in industrial systems (such as message passing and client/server-patterns) must be adequately captured by the timing models. Second, unnecessary pessimism in the analysis must be removed (i.e. the analysis results must correlate well with actual system behavior by providing minimal overestimation). Third, correlated behaviors of different parts of the systems must be accounted for (i.e. not all tasks will experience the worst case execution times at the same time).