TL;DR
Most Scan to BIM errors are inherited from the cloud: SLAM drift in repetitive spaces, no tie to control, coverage gaps modelled from assumption, resolution mismatched to LOD, reflective noise, and registration residuals averaged rather than reported per setup. Verify with a five-minute check — residuals, control tie, a 4' AFF slice, a full-height section, and a coverage-gap list — before modelling begins.
A modeller can only build what the data shows. When a Revit model comes back wrong, the cause is usually upstream: something about the capture, the registration, or the coverage made the correct answer unavailable. Fixing the model does not fix the cause, and the same error reappears on the next floor.
Six defects account for nearly all of it.
1. Drift in long, repetitive spaces
Mobile SLAM capture accumulates error along corridors, tunnels, and warehouse aisles where the geometry gives the algorithm nothing distinctive to hold onto. Walls end up subtly non-parallel, grid lines splay, and the model built from them will not coordinate against a survey.
How it looks: duplicate surfaces a few millimetres apart in overlap regions; a grid that fits at one end of the building and not the other.
The fix, applied in the field: closed loops rather than a chain, and terrestrial setups tying the mobile run back to control. A 250,000 sq ft distribution center held ±6 mm across a live facility using five closed loops and 40 control points. A 220,000 sq ft plant retrofit paired mobile capture with 38 RTC360 setups placed at tie-ins.
2. No tie to project control
A cloud can be internally tight and still be in the wrong place. Without a tie to project or state plane control, the model floats, and every discipline that links to it inherits a different origin. Absolute accuracy to control on our work runs 8–12 mm; on a 480,000 sq ft, twelve-building campus, 47 permanent control points held ±0.02 ft (±6 mm) across 18 acres so all twelve buildings coordinated on one frame.
Cost of finding this late: the model is not wrong internally, so nothing looks broken until a survey or a new addition disagrees with it. That discovery usually happens after design is committed.
3. Coverage gaps treated as geometry
Points missing behind stored product, above hard ceilings, or inside locked rooms are not an accuracy problem — they are a knowledge problem. The failure is when they get modelled anyway, from the drawings, with no annotation. Assumed geometry and measured geometry look identical in Revit.
The rule: unscanned areas get enumerated in the deliverable, in writing. What is not measured is not modelled, or it is modelled and clearly flagged.
4. Resolution mismatched to the deliverable
Capture resolution set for a fast walkthrough will not support LOD 300 modelling of small-bore pipe or connection detail. Set too high everywhere, the dataset becomes unmanageable and the schedule slips for no benefit. Resolution should follow the LOD split — dense where LOD 400 is specified, standard elsewhere. See LOD 200 vs 300 vs 400.
5. Noise from reflective, wet, or dark surfaces
Glass, polished floors, stainless, and standing water throw returns. Untreated, this shows as ghost surfaces behind glazing and speckle above wet slabs, and a modeller working fast will trace it. It is handled at capture — additional setups at oblique angles — and in cleanup, not in Revit.
6. Registration residuals nobody looked at
Interior registration on our work targets 2–4 mm RMS, exterior 4–6 mm, inside a ±5mm registered deliverable. Those are per-setup numbers. Reported as a project average, one bad setup — often the mechanical room, often the most important area — disappears into the mean.
Ask for residuals per setup. An average is a marketing number; the distribution is an engineering number.
What the defects cost downstream
Modelling errors inherited from the cloud surface the same way every other existing-conditions error does: as RFIs and change orders. Renovation projects with unverified conditions run 15–40 RFIs at $1,200–$2,500 each and change orders at 3–8% of contract value, with 10–20 days of schedule slip. Verified capture and modelling runs $0.25–$0.60/sq ft all-in and takes RFIs down to 2–8.
Concretely, on a 220,000 sq ft plant, an LOD 300 model built from properly controlled data exposed 143 pipe runs routed differently than the record PDFs claimed and retired 26 potential RFIs — roughly $340,000 in avoided rework. That value only exists if the cloud is trustworthy enough that the discrepancies are believed.
The five-minute quality check before modelling starts
- Open the registration report. Confirm per-setup residuals, not an average.
- Confirm the tie to control and the coordinate system, in writing.
- Slice a plan view at 4' AFF and look for doubled walls in overlap zones.
- Slice a section through the full building height and check that floors stack.
- Compare the coverage map against the scope and list every gap.
Any of these failing is cheaper to fix with a half-day return visit than with three weeks of modelling on bad input.
Related reading
Point cloud accuracy for BIM coordination covers how these errors propagate into clash tolerance. Point cloud file formats covers what survives the handoff between platforms. For capture methodology, see Building 3D Laser Scanning.