AI in scaffold planning: what works today

What counts as AI in scaffold planning?

The word covers four quite different things, and it helps to keep them apart.

1. Recognition: reading text, scales and outlines from drawings, which is machine learning applied to images.
2. Capture: building a 3D model of the site from a phone's LiDAR sensor or from photographs.
3. Layout: placing bays, lifts, decks and ties automatically. In the tools we checked, vendors describe this as automatic or rule-based rather than as AI.
4. Assistants: large language models that answer questions in plain language, increasingly connected to a company's own data.

The first two are mature and widely available. The fourth is the new part, and it is where most of the recent announcements sit.

How well does drawing recognition work?

A PDF drawing comes in two forms. A drawing exported from CAD keeps its lines and text as data, so a program can read the scale note and follow the walls. A scanned drawing is just a picture, and the text on it has to be recovered by optical character recognition (OCR) before anything can be read. An IFC model goes a step further and carries the building as objects with heights and openings; what that asks of a scaffolding firm is covered in BIM for scaffolding.

The hard part is not reading characters; it is knowing what they mean. A sheet can carry "1:100" for the plan and "1:20" for a detail beside it, dimension strings that refer to the structure rather than the finished facade, and scales that no longer match after a PDF has been printed and rescanned. Good tools therefore only use a ratio that is labelled as the scale, and they ask the user to confirm it against a known dimension.

Recognition of outlines works best on clean plans of simple buildings. Extensions, bay windows and drawings with furniture and hatching confuse it, which is why every serious import still lets you trace or correct the walls by hand. Treat recognised walls as a draft, and check the lengths before they become a parts list.

Can a phone scan replace the tape measure?

Partly. Apple's RoomPlan, built on ARKit, uses the camera and the LiDAR Scanner on recent iPhones and iPads to create a 3D floor plan of a room, including its dimensions (Apple Developer, RoomPlan). It is designed for interiors, so facade work needs other capture methods.

Photogrammetry, which builds a model from many photographs, does not need LiDAR at all. Avontus documents a route in which a phone, tablet or drone running the Scaniverse app captures the building, the mesh is reduced in MeshLab and then imported into Avontus Designer as an OBJ file; the guide says expensive laser or LiDAR scanners are not needed, warns users to watch for obstructed areas, and states no dimensional accuracy for the result (Avontus documentation). That last point applies to every scan: check the key dimensions on site before ordering.

What do vendors claim about AI layouts?

We read the vendors' own pages on 23 October 2026. What stands out is how little of it is called AI.

In other words, the layout engines that place bays and count parts are described as automatic or rule-based, and the products that are called AI are assistants on top of the data. That is a sensible split. Bay layouts, tie patterns and bracing intervals come from manufacturers' rules that can be written down exactly, and a rule engine is easier to check than a learned model.

What can an AI assistant do with scaffold data?

A general chatbot with no connection to your data will answer a scaffold question with plausible numbers, and nothing guarantees they are right. The useful assistants are connected to a tool that does the counting.

Avontus describes its connector as linking its Quantify, ScaffoldIQ, Designer and Viewer data to Claude or ChatGPT, alongside other systems a company connects, and gives an example in which the assistant sorts a pipe-rack scaffold's bill of materials by elevation and splits it into truck loads; Avontus says it does not collect or store the data for AI, the assistant reads each system under the user's existing permissions, and the connector is read-only and cannot create, change or delete anything (Avontus AI Scaffold Connect). ScaffPlan lists skills, agents, more than 100 scaffolding actions, connectors through the Model Context Protocol (MCP), company knowledge and memory, all working inside the customer's rules and approvals (ScaffPlan AI).

The pattern in both is the same: the language model handles the question, and the numbers come from the vendor's own system. When you test any assistant, ask where each number came from. If the answer is "the model", treat it as a guess.

What can AI not do in scaffold planning?

Structural responsibility does not move

An AI cannot be the competent person. In Great Britain, the Work at Height Regulations 2005 require strength and stability calculations for scaffolding unless a note of calculations covering the arrangement is available or the scaffold follows a generally recognised standard configuration, and an assembly, use and dismantling plan drawn up by a competent person (Schedule 3, paragraphs 7 and 8). HSE expects a competent person to inspect the scaffold before first use and at least every 7 days (HSE scaffolding guidance). An assistant can draft, count and remind; it cannot carry that duty.

The EU AI Act timeline

For tools sold or used in the EU, Regulation (EU) 2024/1689, the AI Act, sets the rules. The Commission's page gives this timeline, as amended by the 2026 AI Omnibus, which entered into force on 27 July 2026 (European Commission, AI Act):

For a scaffolding firm two points matter now. Article 4 has applied since 2 February 2025; as amended by the AI Omnibus, providers and deployers must take measures to support the AI literacy of their staff and others who operate AI systems for them, without a prescribed literacy level for any particular person (Cuatrecasas on the AI Omnibus). In practice, staff who use an assistant should know what it can and cannot do. Since August 2026, chatbots have to tell users they are talking to a machine. Whether a particular planning tool falls into a high-risk category depends on how it is used, so ask the vendor how they classify it.

What should you use today?

• Use recognition and scans for speed, then check. Confirm every scale and every key length on the drawing or on site.
• Prefer rule engines for counts. A layout engine that follows the manufacturer's instructions is predictable and can be checked against the manual.
• Use assistants on top of real data. Ask where each number came from, and keep the competent person in the loop.
• Read the vendor's own page. If a feature is "launching soon", it is not yet a feature.

ScaFix, which we make, works the same way: its PDF import runs text recognition in the browser to find a printed scale on scanned drawings, its iPhone app can also scan with LiDAR, and its ChatGPT app, a developer preview at the time of writing, passes wall measurements to the ScaFix engine so that ChatGPT does not invent the numbers (ScaFix in ChatGPT). For how scaffold software works more generally, see what does scaffold software do, and for how scaffold design tools compare, the scaffold software comparison.

This is planning information, not structural approval. An AI tool does not replace the manufacturer's instructions, TG20 or a bespoke design; in Great Britain the Work at Height Regulations 2005 and HSE guidance apply, the EU AI Act governs AI tools placed on the EU market, and a competent person must inspect the scaffold before use.

Frequently asked questions

Sources

Primary sources used for this article, with the date each was read.

• European Commission: AI Act, regulatory framework and timeline (read 23 October 2026)
• Apple Developer: RoomPlan (read 23 October 2026)
• Avontus documentation: real-world buildings into 3D models (read 23 October 2026)
• Avontus: AI Scaffold Connect (read 23 October 2026)
• ScaffPlan: ScaffPlan AI (read 23 October 2026)
• Scaffcalc (read 23 October 2026)
• SMART Scaffolder: SMART Estimator (read 23 October 2026)
• NTI: NTI SCAFFPLANNER (read 23 October 2026)
• Layher: LayPLAN CAD (read 23 October 2026)
• Cuatrecasas: Digital Omnibus on AI has been published (Article 4, AI literacy) (read 23 October 2026)
• legislation.gov.uk: Work at Height Regulations 2005, Schedule 3 (read 23 October 2026)
• HSE: Scaffolding (read 23 October 2026)