In heritage documentation, three-dimensional (3D) models created using Scan-to-BIM processes are essential for interpreting and presenting historic structures. Point cloud data derived from 3D laser scanning and photogrammetry facilitate realistic digital models used for immersive experiences. For this, raw point clouds, which are unstructured, are processed, semantically classified, and segmented to create parametric architectural objects in modeling platforms. ‘Three-Dimensional Point Cloud Semantic Segmentation (3DPCSS)’ refers to segmenting point clouds into classes like walls, columns, etc. Automating 3DPCSS using Artificial Intelligence (AI) has gained importance in current research activities because of its versatility and efficiency over manual segmentation. However, implementing it solely with AI presents various operational and conceptual challenges, particularly for XR models in digital heritage. Automated segmentation often fails to capture the unique characteristics and intricate geometries, leading to misrepresentations or oversimplifications. Selecting an appropriate algorithmic framework for automating 3DPCSS is essential to address this gap. This paper aims to understand the efficacy of AI algorithms in recent research for 3DPCSS, particularly those tailored for 3D modeling. A study of Dwarakadesh Haveli, Ahmedabad, India, highlights the workflow and challenges of integrating point clouds into 3D models. The findings indicate the need for a detailed approach tailored to the project’s specific characteristics, emphasizing the importance of systematic algorithm ensemble experimentation to refine segmentation, leading to the development of 3D parametric objects.