Scan to BIM: A Practical Point Cloud to Revit Workflow

Why scan to BIM keeps landing on your desk

Most of the buildings your firm will touch in the next decade already exist. Renovation, retrofit, extension, and facility upgrades now make up a huge share of AEC work, and every one of those projects starts with the same problem: nobody has a reliable model of the existing building. The drawings on file are decades old, the last renovation was never documented, and the site does not match anything on paper.

Scan to BIM is the workflow that solves this. A laser scanner captures the building as millions of measured points, those scans get stitched into a single point cloud, and a modeler rebuilds the building in Revit on top of that cloud. The output is an as-built model the design team can actually trust.

If you work as a BIM modeler or coordinator, scan to BIM is one of the most employable skills you can add. It is also one of the easiest workflows to do badly. This guide walks through the full pipeline the way it runs on real projects: capture, registration, linking into Revit, modeling method, tolerances, and deliverables, plus the mistakes that quietly ruin scan to BIM projects.

What scan to BIM is, and what it is not

Scan to BIM converts reality capture data into an intelligent building model. The key word is converts. The point cloud itself is not a BIM deliverable. It is a measurement, a very dense one, but it carries no walls, no rooms, no parameters, and no schedulable data. The value appears when a human (with some software help) interprets those points into Revit elements.

Three clarifications save a lot of confusion on projects:

• Scan to BIM is not automatic. Despite years of marketing, no tool reliably turns a raw cloud into a clean, well-structured Revit model without human judgment. Tools like ClearEdge3D EdgeWise or the point cloud features in Revit speed up specific tasks such as pipe extraction or wall fitting, but a modeler still decides what is a wall, what is a bulkhead, and what is junk data from a mirror.
• The model is an interpretation, not a clone. Existing buildings lean, sag, and bulge. A Revit wall is straight and vertical. Every scan to BIM model is a negotiated simplification of messy reality, which is exactly why tolerances (covered below) must be agreed in writing.
• Scan to BIM is not only for heritage projects. The busiest use cases are ordinary ones: office fit-outs, hospital refurbishments, plant rooms before an MEP retrofit, and construction verification where a contractor scans installed work to compare against the coordination model.

How buildings get captured: the hardware landscape

You do not need to operate a scanner to run a scan to BIM project, but you do need to know what the survey team used, because the capture method sets the accuracy ceiling for everything downstream.

Method	Typical accuracy	Speed	Best for
Terrestrial laser scanner (TLS), e.g. Leica RTC360, Faro Focus	2 to 6 mm per scan	Minutes per setup	Structural and MEP as-builts, facades, plant rooms
Mobile / SLAM scanner, e.g. NavVis VLX, Leica BLK2GO	5 to 20 mm	Walk-through pace	Large floorplates, corridors, quick space capture
Drone photogrammetry	10 to 50 mm+	Site-wide in hours	Roofs, facades, topography, inaccessible areas
360 photo capture (no cloud)	Not measurable	Very fast	Visual reference only, not for modeling

A common project setup combines methods: TLS for the plant rooms and structure where millimeters matter, SLAM for the general floor areas, and a drone for the roof. As the BIM professional receiving the data, always ask for the survey report. It should state the equipment used, the registration error, the control network, and the coordinate system. If a survey company cannot produce one, treat their accuracy claims with suspicion.

From raw scans to a registered point cloud

A scanner produces one scan per setup position. A mid-size building might involve 200 to 600 setups. Registration is the process of aligning all of those scans into one consistent cloud, and it is where survey quality is won or lost.

Registration is usually done by the survey team in software like Leica Cyclone, Faro Scene, or Autodesk ReCap Pro. Two things matter to you as the downstream modeler:

1. Registration error. The report should state a mean error, typically a few millimeters for TLS work. If scans are misaligned, you will see it as doubled surfaces: a wall that appears twice, 20 mm apart. Modeling on top of a badly registered cloud produces a model that is wrong everywhere.
2. Coordinate system and control. For anything beyond a single small interior, the cloud should be tied to survey control points in a real-world coordinate system. This is what lets your Revit model, the cloud, and the site agree with each other, and it is what makes shared coordinates work later.

For Revit work, the practical delivery format is RCP (a ReCap project file referencing RCS scan files). If the surveyor delivers E57, a vendor-neutral exchange format, you convert it to RCP by importing it into ReCap. During that conversion, apply decimation sensibly: a 1 cm grid spacing is enough for architectural modeling and makes the cloud dramatically lighter. Keep the original E57 archived untouched.

Getting the point cloud into Revit the right way

Revit consumes point clouds as links, not imports. Use Insert, then Point Cloud, and pay attention to the positioning option:

• Auto - Origin to Origin places the cloud’s origin at Revit’s internal origin. Fine for a small standalone interior scan.
• Auto - By Shared Coordinates is the correct choice on any georeferenced project. Acquire coordinates properly so the cloud, the survey, and every linked model agree.

Never move the point cloud to suit your model. Set up your Revit project (project base point, true north, levels) to suit the cloud and the survey control. The cloud is the measurement; your model is the thing that must conform.

Performance housekeeping makes or breaks the modeling experience:

• Keep the RCP and RCS files on a fast local drive or a properly synced cloud workspace. Point clouds over slow network drives make Revit crawl.
• Control visibility per view. Point clouds have their own category under Visibility/Graphics. Turn them off in every view that does not need them, especially sheets and schedules views.
• Use section boxes aggressively in 3D views. Revit only renders the points inside the box, which keeps orbiting smooth even with clouds of several hundred gigabytes on disk.
• Consider a dedicated “cloud tracing” 3D view and a set of narrow plan and section views with tight view ranges, rather than switching the cloud on and off in your main working views.

A working method for modeling from the cloud

Modeling from a point cloud is a discipline of its own. The order of operations below is the one that holds up on real projects.

1. Establish levels and grids first. Cut sections through the cloud and set your levels from the actual slab surfaces, averaged sensibly per floor. If the project has structural grids, derive them from column centerlines in the cloud. Every element you model afterwards references this skeleton, so spend real time here.

2. Model structure before architecture. Columns, slabs, load-bearing walls, and beams give you the frame that everything else hangs on. In plant rooms, model primary steel before chasing pipes.

3. Trace walls in plan at a consistent cut height. Set a plan view range that cuts around 1.2 m above finished floor, thin the cloud display, and trace. Where a wall varies in thickness along its length, pick the dominant thickness and note the deviation rather than modeling every bulge.

4. Use sections constantly. Plan tracing alone lies to you. Cut a section every few meters to check that the wall you traced in plan is actually plumb enough for a single Revit wall, and to pick up soffit steps, bulkheads, and slab falls.

5. Model MEP from the cloud only where scoped. Extracting pipework, ductwork, and cable containment from a cloud is slow, detailed work. If the scope says “visible services in plant rooms at LOA 30,” model exactly that. Do not freelance an entire building’s services because the points are there.

6. Verify as you go. Revit shows points and model together, so make a habit of orbiting with a section box and visually checking model against cloud every session, not at the end. Dedicated verification tools such as Verity can produce deviation heat maps for formal QA on larger projects.

Tolerances and Level of Accuracy: agree the numbers first

The single most important document on a scan to BIM project is the one that states how accurate the model must be. The industry reference is the USIBD Level of Accuracy (LOA) specification, which defines accuracy bands and, critically, separates the accuracy of the measurement from the accuracy of the representation.

LOA	Accuracy range (2 sigma)	Typical use
LOA10	5 cm to 15 cm	Massing, space planning, early feasibility
LOA20	15 mm to 5 cm	General architectural as-builts, fit-out design
LOA30	5 mm to 15 mm	Coordination-grade as-builts, most MEP retrofits
LOA40	1 mm to 5 mm	Fabrication interfaces, heritage recording
LOA50	0 to 1 mm	Specialist metrology, rarely used in AEC

Two practical implications:

• Represented accuracy costs modeling time, not scanning time. The scanner already captured the building at millimeter level. Asking for LOA40 representation means the modeler must chase every deviation, which can triple the modeling budget compared to LOA20.
• Different systems can carry different LOA. A sane specification might say LOA30 for structure and plant rooms, LOA20 for architectural walls and openings, and LOA10 for the roofscape. Pricing and planning become much easier when the spec is granular.

Pair LOA with Level of Development (LOD) in the BIM execution plan: LOD tells you how much element data and detail to model, LOA tells you how faithfully it must sit against reality. They answer different questions and a scan to BIM brief needs both.

Old buildings are not straight: handling deviation

Sooner or later you hit the philosophical problem of scan to BIM: the cloud shows a wall that leans 40 mm over its height, a slab that falls 60 mm across a room, a colonnade where no two columns match. Revit wants ideal geometry. What do you do?

The professional answer is a deviation strategy, agreed with the client before modeling starts:

• Within tolerance: model ideal. If the lean is inside the agreed LOA band, model the wall vertical at the mean position and move on.
• Outside tolerance but locally: model ideal, flag it. Place the element at best fit and record the deviation, either in a shared parameter on the element or a tracked issue list. The design team needs to know about the 40 mm lean; they do not need a sloped wall that breaks every downstream dimension.
• Structurally meaningful: model true. Sagging beams, out-of-plumb facades on a facade retention job, or anything a structural engineer will analyze should be modeled as measured, often with in-place or adaptive components.

Whatever you choose, write it down. The fastest way to lose a client’s trust is for them to discover, mid design, that the model silently idealized something they were relying on.

Scope and deliverables: the conversation before the modeling

Scan to BIM projects go wrong at the briefing stage more often than in Revit. Before anyone models a wall, the following should exist in writing, usually inside the BIM execution plan:

• Element scope list. Exactly which categories get modeled: walls, doors, windows, structure, visible MEP, sanitary fixtures, ceilings, and so on. “Model everything” is not a scope.
• LOA and LOD per system, as covered above.
• Areas in and out of scope. Ceiling voids? Risers? Roof plant? Under-floor? A scanner cannot see through a suspended ceiling, so concealed services either need opened access panels during capture or an explicit exclusion.
• The unknowns clause. Some spaces will be locked, occupied, or unsafe on scan day. Agree how gaps are handled: rescan visit, model from drawings with a “not verified” tag, or leave void.
• Deliverable format. The Revit version, whether the point cloud itself is delivered, the coordinate system, and naming per your file standards. If the client works to ISO 19650, the as-built model and the cloud are separate information containers with their own status codes.

A one-page scope table agreed by email beats a week of rework every single time.

Common mistakes that ruin scan to BIM projects

These are the failures that show up repeatedly in audits of scan to BIM models:

1. Modeling from an unverified cloud. The team starts tracing on day one and discovers registration errors in week three. Always review the registration report and spot-check known dimensions first.
2. Ignoring shared coordinates. The cloud sits at origin, the model sits somewhere else, and every consultant who links the file inherits the mess. Set coordinates up on day one.
3. Over-modeling. The modeler faithfully rebuilds every pipe bracket at LOA40 when the client is paying for an LOA20 fit-out model. The budget evaporates and nobody benefits.
4. Under-communicating gaps. Areas the scanner never saw get quietly modeled from old drawings without any flag. Six months later the contractor finds a wall 300 mm from where the model claims.
5. One giant cloud in every view. Revit performance collapses, the team blames the software, and productivity dies. Manage visibility and use section boxes.
6. Treating the cloud as disposable. The RCP gets deleted after modeling to save disk space, then a query lands about a dimension nobody modeled. Archive the cloud with the project; it is the evidence behind the model.
7. No QA pass against the cloud. The model ships without anyone systematically checking model-to-cloud deviation. Even a manual sampling check per floor catches most howlers.

Best practices checklist

If you want a compressed version to pin next to your monitor:

1. Demand the survey and registration report before opening Revit.
2. Convert to RCP, decimate to around 1 cm for architectural work, archive the raw E57.
3. Link the cloud by shared coordinates and never move it.
4. Build levels and grids from the cloud first, then structure, then architecture, then scoped MEP.
5. Trace in plan, verify in section, orbit with a section box every session.
6. Model to the agreed LOA, flag deviations beyond tolerance, and record the deviation strategy.
7. Tag unverified areas honestly.
8. Run a model-to-cloud QA pass before delivery, automated on large jobs, sampled on small ones.
9. Deliver model, cloud, and a short methodology note as separate, named information containers.

Where scan to BIM fits in your BIM career

Scan to BIM sits at a useful intersection: it is technical enough that most generalist modelers avoid it, and in demand enough that firms pay a premium for people who can run it end to end. Renovation-heavy markets, including most of Europe, have more existing-building work than new build, and every one of those projects needs someone who can turn a cloud into a model the team can build on.

The skills stack neatly on what you may already have. Strong Revit modeling fundamentals carry most of the weight, and the additions are learnable: reading registration reports, setting up shared coordinates correctly, working to a written LOA specification, and developing the judgment to know when reality is close enough to ideal geometry.

If your Revit fundamentals need reinforcement first, the structured courses at Archgyan Academy cover the modeling, coordination, and standards workflows that scan to BIM builds on, taught from real project practice.

Start small if you can: volunteer for the next existing-building project in your office, sit with the survey team for a day, and model one plant room properly against a cloud. One well-run scan to BIM project on your record says more than any software certificate.