The reconstruction of as-built building models from point cloud data is a challenging problem with promising applications in the building operation and maintenance (O&M) phase. As-built models are more accurate than as-designed ones and thus enable a better understanding and decision-making regarding retrofitting buildings to changing user needs, or making them compliant with regulations. Currently, the model reconstruction task involves significant manual modeling effort. In order to achieve improved automation of as-built building model reconstruction, we investigate the derivation of missing semantic data. These include topological relations between objects in a building (e.g. access relation between spaces) and semantic object properties (e.g. space functions). We propose a model reconstruction pipeline which is divided into point cloud processing, topology reconstruction, and space classification steps. Methods are developed to: i) extract semantically labeled geometric models from a point cloud, as required by O&M applications, ii) reconstruct the geometry of building components (e.g. walls) and spaces, including their topological relations, and iii) enrich space and related object data with functional classification data. We reuse, adapt, or extend known algorithms and data structures from the areas of point cloud processing, topology modeling, and semantic model enrichment. A proof of concept of the reconstruction pipeline is implemented and validated for O&M application domains of room area measurement and evacuation path analysis.