Architecture-based adaptation equips a software-intensive system with a feedback loop that enables the system to adapt itself at runtime to changes to maintain its required quality goals. To guarantee the required goals, existing adaptation approaches apply exhaustive verification techniques at runtime. However these approaches are restricted to small-scale settings, which often limits their applicability in practice. To tackle this problem, we introduce an innovative architecture-based adaptation approach to solve a concrete practical problem of VersaSense: automating the management of Internet-of-Things (IoT). The approach, called MARTAS, equips a software system with a feedback loop that employs Models At Run Time and Statistical techniques to reason about the system and adapt it to ensure the required goals. We apply MARTAS to a building security case system, which is a representative IoT system deployed by VersaSense. The application comprises a set of IoT devices that communicate sensor data over a time synchronized smart mess network to a central monitoring facility. We demonstrate how MARTAS outperforms a conservative approach that is typically applied in practice and a state-of-the-art adaptation approach for different quality goals, and we report lessons learned from this industrial case. © Springer Nature Switzerland AG 2018.