Scan-to-BIM Workflow for Existing Buildings

In existing buildings, the drawings are never right. Anyone planning a refurbishment, an extension or an energy retrofit knows the pattern: the 1978 record drawing shows a wall where today there is a steel beam and a ventilation shaft. This is exactly where scan-to-BIM earns its place. A LiDAR scan becomes an as-built BIM model that architects, MEP designers and structural engineers can actually plan with. Below is the workflow step by step — and what you, as the client, need to hand to the provider so the result is worth the money.

Why scan-to-BIM at all?

Three reasons that justify almost every scan-to-BIM brief in practice:

1. Refurbishment & conversion: existing drawings are often decades old, incomplete or simply wrong after several rounds of alterations. A BIM model from a point cloud delivers a verified as-built instead of guessed dimensions.
2. Extensions and over-builds: the moment new construction docks onto existing fabric, structural engineers and MEP need exact connection points. An as-built BIM model is the foundation for clash-free design.
3. Energy retrofit: energy audits, funding applications (BEG, KfW in DE; PAS 2035 in the UK) and renovation roadmaps increasingly demand a digital building model — no longer just a PDF.

Whoever plans in the existing fabric without a current survey plans twice: once at the desk, once after the first wall has been opened on site. The scan-to-BIM workflow moves the surprises out of the construction phase and back into CAD.

The workflow in five steps

Our standard workflow for scan-to-BIM projects runs in five clearly separated phases. Each phase has its own tools, its own deliverables and its own acceptance criteria.

1. Capture — LiDAR + photogrammetry

For existing buildings we typically combine a terrestrial LiDAR scanner (Faro Focus, Leica BLK360 or equivalent) for the metric backbone with photogrammetry or Matterport Pro3 for visual texturing and quick walk-through documentation. Per setup, typical accuracies sit at ±2–4 mm over 10 m. Important: the outcome is decided already here — a coarse scan grid makes walls fuzzy further down the line.

2. Registration & cleaning

The individual scan setups (often 80–300 per floor) are registered into a single coordinate system — by cloud-to-cloud matching, measured targets or control points. Then comes the cleaning: people, moveable furniture, reflections in glazing and scan artefacts are removed. The result is a consolidated E57 or LAS point cloud — the actual raw material for the BIM model.

3. Classification — floor, ceiling, wall, column, opening

Now the point cloud receives semantics. Using tools such as CloudCompare, Autodesk ReCap Pro, Leica Cyclone or specialised AI classifiers (PointCAB, Faro As-Built), point groups are assigned to component categories: load-bearing wall, partition, floor, ceiling, column, door, window, parapet, balustrade. This is what stops the next step turning into hand-tracing every wall from scratch.

4. Mesh & geometry extraction

From the classified point cloud, geometric primitives are extracted — planes for floors and ceilings, axes for columns, boundary lines for walls and openings. Where needed (ornate ceilings, vaults, curved surfaces), an additional mesh is generated. This is the step where you discover what is cleanly automatable (orthogonal office buildings) and what stays manual (period buildings, listed substance).

5. Modelling in the BIM tool — LOD 200, 300, 400

Only now does actual BIM authoring begin. The point cloud is loaded as a reference inside Revit, ArchiCAD or an OpenBIM tool, and parametric components are placed against the extracted geometry: wall with layer build-up, slab with material, window with reveal, door with swing direction. Delivery is typically native (.rvt, .pln) plus IFC 4 for vendor-neutral exchange.

Level of Detail — LOD 100 to 400

The LOD level fixes how precisely each component is modelled in the as-built BIM model. It is the single most important specification you, as the client, decide. Higher LOD = more hours = more expensive.

For the great majority of existing-building projects, LOD 200 or LOD 300 is the economic sweet spot. LOD 350 pays off when MEP and structure are being coordinated in parallel. LOD 400 almost never — unless specific pre-fabrication is on the table.

What we need from the client

A good scan-to-BIM quote starts with a good brief. When you request a quote — from us or anyone else — clarify these five points up front:

1. Target LOD per discipline: architecture LOD 300, MEP LOD 200, structure LOD 350? A blanket "LOD 300 across the building" is usually more expensive than necessary.
2. Target software format: Revit (.rvt), ArchiCAD (.pln), Allplan (.pln), or pure IFC exchange? Anyone continuing to design in Revit wants native Revit families, not generic IFC walls.
3. Level of detail at doors and windows: generic standard door, or with ironmongery, glazing type, U-value? For energy retrofits the second one is mandatory; otherwise you can save the effort.
4. Connection standards: which family libraries (office standard, your own templates, VDI 2552 in DE, NBS in the UK), which classification system (Uniclass, OmniClass, your own)?
5. BIM Execution Plan (BEP): if a BEP exists, send it with the brief. If not, we settle the minimum points in a 30-minute call before quoting.

Rule of thumb: the more precisely you specify here, the less risk premium ends up in the quote. A "we just want some BIM" easily costs 50% more than a precise brief — because the provider has to price in the worst case.

Software pipelines we use

We are tool-agnostic — what matters is what your design team works with afterwards. Our most common pipelines:

• Revit pipeline: Leica Cyclone / ReCap Pro → Revit + PointSense / As-Built for Revit. Output: .rvt + IFC 4.
• ArchiCAD pipeline: CloudCompare → ArchiCAD with BIMx cloud tools. Output: .pln + IFC 4.
• OpenBIM: pure IFC delivery for clients with their own tooling, coordination via BIMcollab or Solibri.
• Structural extension: on request, export to Tekla or SCIA for analytical models.

Common pitfalls

What we see again and again on existing-building projects — and how we avoid it:

• Point cloud too dense: an untrimmed 500-million-point cloud will crash even good workstations. We decimate task-specifically (architecture lands around 10–20 million points per floor).
• Hidden surfaces: what the scanner did not see (behind cupboards, in service shafts, above suspended ceilings) does not exist in the cloud. Clarify before the scan: which components must end up in the model? Moving cupboards is worth it.
• Classification errors: AI classifiers occasionally read a bookshelf as a wall. Without human QA, that ends up in the model.
• Reflections and glass: glazed surfaces reflect LiDAR partially or not at all. Window areas often have to be reconstructed manually — effort that belongs in the quote.
• Coordinate-system drift: when the scan and the existing drawings sit in different references (local vs. UTM, Swiss LV95 vs. CH1903), the transformation gets forgotten. Clarify up front, no drama later.
• Component hierarchy undefined: without a clear rule on whether "partition" lives in one family or as a storey-height divider, the Revit model ends up with double work later on.

When is a plain point cloud enough?

Not every existing building needs BIM. Be honest with yourself before commissioning:

• Who works with the model afterwards? If all that's needed in the end is a PDF floor plan, BIM is overkill. Point cloud + 2D plan is entirely enough.
• What CAD does the designer use? If the architect lives entirely in AutoCAD LT, an IFC model is useless to them — point cloud + 2D is the right answer.
• Is it visualisation or design? For marketing, virtual viewings and as-built documentation, a Matterport twin plus a point cloud is often enough. BIM only becomes necessary when MEP, structure or building physics join the party.
• Will the model be maintained? A BIM model that is not maintained after handover is out of date within a year. For pure as-built documentation without a maintenance concept, point cloud + snapshot is often the smarter call.

Rule of thumb: under 300 m² of office area, or on purely architectural projects, BIM modelling often costs more than it saves. On energy retrofits above 800 m², on extensions and over-builds, and on MEP-heavy refurbishments, scan-to-BIM almost always pays.

Conclusion

A properly run scan-to-BIM process turns an existing building into a plannable asset. The point cloud is the metric truth; the BIM model is the usable translation. If you are planning a refurbishment, an extension or an energy retrofit in existing fabric, BIM from a point cloud is usually the cheaper option — measured against the follow-on costs of wrong as-built assumptions on site.

We will give you an honest opinion on your specific existing-building project — including the case where a BIM model is not the right lever for your task: Request consultation.