Automatic Scan-to-BIM involves 3D modelling of buildings from lidar data. The modelling involves registration, noise removal segmentation, classification and then reconstruction in 3D. These steps are becoming automated with less need for expert supervision. However, among them, the reconstruction step is less automated and heavily relies on human operators. This paper identifies and classifies common errors, including completeness, geometrical and topological errors, in the automatic reconstruction of building indoor models. It then formulates deterministic solutions for different error types. The solutions are then integrated and applied. The qualitative result analysis indicates that the deterministic solutions, to a large degree, are effective in removing errors at lower levels of detail. This paper signifies deterministic methods as a practical alternative to facilitate automatic Scan-to-BIM.