From Laser Scan to Revit As-Built Drawings: What AI Actually Does (and Doesn't) on Site

As-built drawings are one of those jobs everyone in practice dreads and nobody can really avoid. An existing building has to be measured, checked and redrawn accurately before you can design an addition, a renovation, or even confirm what’s actually there versus what the old plans claim is there. Done by hand, it’s slow — days of drafting for a single house isn’t unusual. A handful of companies in South Africa now sell scan-to-BIM as a service, converting a laser scan into a coordinated model for a fee. I’ve gone a different way: building the workflow myself, using AI to draft the first pass, and staying honest that it’s still early.

The old way vs the scan-to-drawings workflow

The traditional route is: measure on site (tape, laser distance meter, or a full survey), then redraw everything by hand in CAD or Revit, checking dimensions as you go. It works, but it’s slow, and mistakes creep in over the hours of manual re-measurement and redrawing.

The workflow I’ve built instead starts with a handheld LiDAR scanner on site, which captures a point cloud of the existing building. Then I talk to the AI, describing what I need, and the system drafts a Revit model and a set of as-built drawings from the point cloud, which I then check and correct against the scan. It’s not scan-to-BIM as an outsourced service; it’s a workflow inside my own practice, built so the first pass is fast and the review is where my time actually goes.

Site context feeds into this too. Where the property already has a Zonely site plan set up — plot boundary, existing footprints, setbacks, height limits and coverage from the zoning scheme — that comes into Revit already in place, so the scan-derived model sits on a site that’s already correctly set out. See the Zonely case study for how that connection works.

What happened on my first project

The first live job for this workflow was a house in Pinelands, Cape Town, in September 2026. By my own estimate, the initial model and drawing generation from the scan took about 30–45 minutes. Review and correction are separate, additional time — I’m still validating the complete workflow end to end, so this is an early result from one project, not a promise about how it performs on every job or every building type.

The system also prepares images to help with that review: point cloud shown behind the drawn walls, so I can see at a glance what’s been drafted correctly and what needs a second look. Anything doubtful gets flagged for me rather than filled in — that’s a deliberate rule, not an accident of how the tool happens to behave. The model also exports to CAD, for consultants on the job who don’t work in Revit.

A checklist before you try this

If you’re considering a scan-based as-built workflow, whether built by you or bought as a service, this is what I’d check first:

1. Match the scanner to the job. A handheld LiDAR scanner is fast and portable but trades some accuracy for speed compared to a tripod-based terrestrial scanner — know which one the job actually needs.
2. Register the scan properly before anything else happens. Errors here propagate into every wall and dimension that follows.
3. Treat the first model as a draft, always. Whether AI or a person drafts it, nothing goes to a client or a builder unchecked.
4. Budget real time for review and correction, separate from generation time. On my numbers so far, generation is the fast part — checking it is where the real time still goes.
5. Decide what “good enough” means before you start. As-built accuracy for a renovation brief is a different bar to as-built accuracy for a structural assessment.
6. Keep the point cloud. It’s the one source of truth you can check any drawn element back against, at any point later in the project.

Limits — what still needs checking

I want to be specific rather than promotional here, because this is the newest workflow I run and the one with the least track record. It’s had one live project. The 30–45 minute figure is my own estimate from that one job, and it covers initial generation only — not the review and correction that follows, which takes real, separate time and is where accuracy actually gets confirmed. I’m not presenting this as a finished product or a guaranteed time saving for any given building; it’s an early result I’m continuing to test, and I’d rather say that plainly than oversell it.

The workflow also doesn’t replace judgement about what’s actually on site. Existing non-standard construction, hidden structural elements, and anything that isn’t visible to a laser scanner — inside a wall cavity, for instance — still needs a person who knows what they’re looking at.

If as-built accuracy matters because you’re planning a renovation or extension, it’s worth also checking what you’re allowed to build on the site before drawings go too far — an accurate as-built is only useful once you know what the site actually permits.

Where this fits in a bigger workflow

Scanning and drafting is one piece of a larger pattern I use across the practice — get accurate information into the project early, let AI handle the first pass on the repetitive parts, keep a person checking before anything is final. The other pieces — project setup, fee proposals, council forms — are described at the scan-to-drawings case study and across the rest of the case studies, and how a build like this gets scoped for another firm is at Services.

If you’re weighing up whether a scan-based workflow is worth building for your own practice, book a free 30-minute call and I’ll tell you honestly where it would help and where it’s still too early to rely on.