Research, in plain language

Decide this before you scan

November 5, 2026 · Bryan Nelson, Co-Founder · 6 min read

Most problems with an as-built Revit model start before the scanner leaves the truck. That is the common thread in fifteen years of academic work on scan to BIM. Below are four widely cited studies, what the authors found, and what it means when you order an existing conditions model for a remodel or addition.

1. Point cloud to Revit is still skilled work

In 2010, a team from Carnegie Mellon and the National Institute of Standards and Technology reviewed the methods researchers had tried for building as-built models automatically from laser scans (Tang et al., 2010). They split the job into three parts: modeling the geometry, recognizing what each object is (a wall, a door, a beam), and working out how the objects relate to each other. They concluded that the work was mostly manual, slow and prone to error, and that all three parts needed more research.

Software has improved since then, and newer tools can find walls and floors on their own. But the step that decides whether a model is useful, judging what each surface is and how it should be drawn, still relies on a trained modeler.

2. Existing buildings are where BIM is weakest

A 2014 literature review from the Karlsruhe Institute of Technology looked at how BIM was being used for buildings that already stand (Volk, Stengel and Schultmann, 2014). The authors found that BIM was well established for new design, but rarely used for existing buildings. The barriers they named were the effort of capturing and modeling what is there, the uncertainty about hidden or undocumented conditions, and the work of keeping a model current once it exists.

3. Start with what the model is for

The most practical of the four is a 2019 framework published in the journal Remote Sensing (Wang, Guo and Kim, 2019). It sets out four steps, in this order:

1. Information requirements. What decisions will the model support, and which elements and level of detail does that need?
2. Scan data quality. From those needs, set the accuracy, point spacing and coverage the scan must reach.
3. Data acquisition. Plan scanner positions and settings to meet that quality, then scan.
4. Model reconstruction. Build the model to the requirements set in step one.

The point is the order. Teams that scan first and ask questions later either over-collect (paying for detail nobody uses) or under-collect (sending someone back to site).

4. Check the whole model against the scan

A 2013 study from Carnegie Mellon looked at how to check the quality of an as-built model made from scan data (Anil et al., 2013). The common practice was to spot-check a handful of dimensions. The authors instead compared the whole model with the point cloud and colored every surface by how far it sat from the scan. In their case study, this deviation analysis found errors that spot checks missed, such as walls in the wrong place or modeled at the wrong size.

Put the accuracy in writing

The industry now has a shared way to state the accuracy that step two asks for. The U.S. Institute of Building Documentation (USIBD) publishes a Level of Accuracy (LOA) specification that sets ranges from LOA10 (coarsest) to LOA50 (finest), and separates the accuracy of what was measured from the accuracy of what was drawn. Our default is a point cloud at LOA30 and an architectural Revit model at LOA20, with LOA40 available on critical elements when the job has survey control.

If you remember one thing from these papers, make it this: a good set of as-built drawings starts with a written answer to "what is this model for?" The scanner, its settings, the modeling hours and the price all follow from that answer.

Sources

1. Tang, P., Huber, D., Akinci, B., Lipman, R. and Lytle, A. (2010). Automatic reconstruction of as-built building information models from laser-scanned point clouds: A review of related techniques. Automation in Construction, 19(7). doi:10.1016/j.autcon.2010.06.007. Free copy: NIST.
2. Volk, R., Stengel, J. and Schultmann, F. (2014). Building Information Modeling (BIM) for existing buildings: Literature review and future needs. Automation in Construction, 38. doi:10.1016/j.autcon.2013.10.023.
3. Wang, Q., Guo, J. and Kim, M.-K. (2019). An application oriented scan-to-BIM framework. Remote Sensing, 11(3), 365. Open access (CC BY): doi:10.3390/rs11030365.
4. Anil, E. B., Tang, P., Akinci, B. and Huber, D. (2013). Deviation analysis method for the assessment of the quality of the as-is Building Information Models generated from point cloud data. Automation in Construction, 35. doi:10.1016/j.autcon.2012.10.005.
5. U.S. Institute of Building Documentation. Level of Accuracy (LOA) Specification. usibd.org.

Summaries are our own reading of each paper. Quotes are not used; see the original papers for full methods and results.

← All posts