Will BOM Replace the Quantity Surveyor?

Earlier this week I received a notification advertising bills of materials prepared for simple houses at a price so low, that no QS can compete with. It seems that the firm uses AI to extract the BOM.

Is AI starting to replace the QS?

I asked the opinion of a senior partner in a large firm, and his answer was reassuring: “We have been using models for a while to extract quantities to produce BoQ. The limiting factor is that the extracted quantities are only as accurate as the model. If the model is perfect then one can get all the quantities 100%, but to date we have not received a perfect model. The QS will still be necessary to check and map the model. The computer or model will also never be able to do cost control or cost management.”

My mind was cast back to the early days when my small practice was doing quite well and invested in a Philips word processor. I did work with an architect’s firm in Sasolburg. They had branches all over SA with the largest branch in Johannesburg.

They had just invested in a computer system that cost them a small fortune in 1983/4!

A senior partner invited me to coffee in Johannesburg. His words still ring true: “Let me show you the future and the end of your profession.”

His pride and joy was a machine using multiple 10” floppy disks. An operator/architect was designing on a large amber (or was it green?) monitor. Next to the gentleman was a library of files containing codes. Every item had a code. Very impressive.

I asked the operator to show me how he could move an external door from a room with a plaster and face brick wall to a face brick wall both sides.

We had two cups of coffee, and the operator was still struggling and stressed.

I drove back to Vanderbijlpark knowing it will take some time before our profession disappears.

Back to the present.

AI models demonstrate high and rapidly increasing efficiency and accuracy in producing Bills of Materials (BOMs) from architectural and engineering drawings, though they still face challenges that require human oversight.

The core of AI's effectiveness in this task lies in its ability to automate processes that were previously manual, time-consuming, and error-prone.

Efficiency of AI in BOM Generation

AI systems offer significant efficiency gains over traditional methods:

Speed: AI-based systems can scan, analyse, and extract data from CAD files, blueprints, and other technical drawings in a fraction of the time it takes a human operator—often reducing analysis time from hours or days to minutes.

Automation: AI uses Computer Vision for symbol/component recognition and Optical Character Recognition (OCR) and Natural Language Processing (NLP) to extract and understand text, such as part numbers, descriptions, and specifications. This automation eliminates manual data entry and the associated bottlenecks.

Workflow Integration: AI-generated BOMs can be seamlessly integrated with ERP (Enterprise Resource Planning), MRP (Material Requirements Planning), or PLM (Product Lifecycle Management) systems, which further accelerates procurement, planning, and cost estimation.

Accuracy of AI in BOM Generation

The accuracy of AI models is generally high, especially with modern, purpose-built systems, but it is context-dependent:

Specialised Models: Highly trained deep learning models, like those based on Mask R-CNN for image segmentation, have shown high mean average precision (mAP) (e.g., up to 98% in some specific component identification tasks) in identifying, classifying, and extracting specific components like concrete formwork.

Error Reduction: By automating the data extraction and cross-referencing process, AI significantly reduces human errors such as missing components, incorrect quantities, or mismatched specifications that are common in manual data entry.

Validation and Consistency: AI can be programmed to cross-reference the extracted data against established standards, databases, and even other drawing elements (like grid lines and title blocks), ensuring greater consistency and identifying potential design mistakes early.

Current Challenges and Limitations

Despite the advancements, AI is not yet a perfect replacement for human review:

Data Fragmentation and Format Variability: While AI is improving in handling diverse formats (PDFs, scanned images, various CAD files), a major challenge remains the fragmentation and inconsistency of product data. BOMs living in static spreadsheets or across disconnected systems can confuse AI, as the models struggle to infer the semantic relationships and context of the components without structured data.

Complex or Novel Designs: General large AI models can struggle with complex shapes, irregular layouts, or non-standard annotations. Specialised models, while more accurate, require extensive labelled, self-developed datasets to perform well, which is resource-intensive to create.

Need for Verification: Due to the potential for "hallucinations" (confident-sounding but incorrect results) and missed detections, industry practice strongly recommends establishing a closed-loop process involving manual spot-checks and verification of the AI-generated BOM to ensure accuracy, especially for critical elements related to payment or safety.

Contextual Understanding: AI is powerful in what it sees and what the text says, but it can still lack the deep contextual and design intent understanding of an experienced engineer or architect, making professional judgement necessary for final sign-off.

Contribution: Bert van den Heever