No more train delays. That is the grand vision of Predictive Maintenance (PM): using data analytics to predict potential failures so well, that these can be prevented by just-in-time maintenance: performing maintenance exactly when and where needed. Even though this grand vision may never be realized, predictive maintenance is widely seen as a very effective maintenance strategy, and as one of the most promising applications of the Internet-of-Things. McKinsey expects cost savings upto $630 billion and down time reductions by 50%. Realizing the envisioned benefits is, however, nontrivial. Companies experience major obstacles in leveraging predictive maintenance technologies. Individual building blocks for PM exist, however, leveraging these into optimal asset management solutions requires major challenges to be tackled. In this talk, I will elaborate on how software engineering lies at the heart of many challenges in predictive maintenance, especially when it comes to automation of the solutions, their specification and verification, multi-stage optimization, the complexity of IT architecture, as well as the organizational embedding of the solutions. Thus, I hope to inspire computer scientists to contribute to this relevant and exciting field. Moreover, I will highlight how formal methods can contribute to this existing field.