Every new edition of NIME brings dozens of new DMIs and the feeling that only a few of them will eventually break through. Previous work tried to address this issue with a deductive approach by formulating design frameworks; we addressed this issue with a inductive approach by elaborating on successes and failures of previous DMIs. We contacted 97 DMI makers that presented a new instrument at five successive editions of NIME (2010-2014); 70 answered. They were asked to indicate the original motivation for designing the DMI and to present information about its uptake. Results confirmed that most of the instruments have difficulties establishing themselves. Also, they were asked to reflect on the specific factors that facilitated and those that hindered instrument longevity. By grounding these reflections on existing reserach on NIME and HCI, we propose a series of design considerations for future DMIs.