Is Your BIM Model Ready for AI?

Introduction

A significant digital transition has been witnessed in the Architecture, Engineering, and Construction (AEC) industry over the past 15 years. During this period, Building Information Modelling (BIM) has proven to be the most crucial component contributing to this advancement. BIM, which started as a basic approach for 3D digital modelling, is now deeply rooted in the industry. By means of BIM, stakeholders and professionals can now manage and share information seamlessly, ensuring overall project efficiency. BIM adoption in today’s industry represents a major breakthrough. Figures indicate that 72.3% of industry experts have adopted it and 15.7% are in the process of its implementation. This fact makes it evident that BIM is no longer a choice but a necessity to excel and help shape the industry’s future.

As time progressed, BIM received mainstream acceptance in the construction industry. While doing so, another technology emerged and gained the spotlight, known as Artificial Intelligence (AI). Upon comparison of the 2020 and 2023 stats, it can be stated that there was a profound change in AI utilisation by the construction industry in its daily routine. The stats show 9% and 22%, respectively, which is a notable change. By 2025, the number rose to 42.5%, with 37.7% of the industry planning its adoption in the upcoming years. When AI surged into view, many people were uncertain whether it was an experimental or a work automation tool. However, later on, this chaotic situation turned into certainty, and companies started focusing on HOW and WHEN to adopt it rather than WHY. This is exactly what the figures show.

88.8% of experts affiliated with the construction industry state that AI, within the coming two to three years, will have a significant impact and change the way they work. However, the changes will be positive, such as improvements in productivity (88.8%), increased safety (72.6%), support for sustainability efforts (69.3%), and increased project reliability (64.5%).

Tasks such as identifying design errors, predicting equipment maintenance, and generating design options already incorporate AI, ultimately making this a practical tool of utmost importance. However, what makes its output stand out is actually the data it is trained on. Well-known AI tools such as ChatGPT (86.8%), Microsoft Copilot (57.9%), and Google Gemini (36.8%) are trained on high-quality data. They require good data from BIM models to produce reliable results. If this condition is not met and AI is trained on inconsistent standards and poor BIM models, AI will fail to produce the expected results. In short, to acquire desired outcomes, it is necessary for organisations to understand the importance of standardised, high-quality, data-rich BIM models.

Requirements for AI

Artificial Intelligence (AI) can perform to the best of its ability as compared to humans, whether it is processing enormous amounts of information or detecting minor errors and omissions. However, unlike humans, AI encounters some constraints; that is, it works on logic and rules, which, in turn, increases the importance of organised data in a BIM model for seamless analysis and interpretation. Data should be structured into the following four core elements.

1. Consistent Parameters

AI primarily operates on pattern recognition and cross-referencing. If the parameters lack these, AI will get confused and it becomes overwhelming to produce precise results. So it is essential to ensure that information is stored in the same way everywhere using consistent parameters.

For instance, if an estimator assigns AI the task of detecting and listing all wall materials, then it is important that the material is stored in a consistent parameter, which could be anything. However, the case should not be like the one below:

• Wall 1: Material
• Wall 2: Wall Type
• Wall 3: Material Info

The above parameter inconsistencies could lead to data not being recorded or becoming fragmented, which ultimately results in poor outcomes. In short, there should only be one parameter, either Material or Material Info. Having two creates chaos.

2. Structured Data

Structured data means every parameter is consistent, and the values in those fields are stored in a properly mapped format. If the data is incorrectly listed in the fields, it may lead to misinterpretation when analysed by AI. For instance,

• Wall 1: Height: 2m, Material: Concrete, Cost: 5000
• Wall 2: Height: 2m, Material: Conc, Cost: 5000

Even though both walls have the same height, material, and cost, if the material “Conc” is not mapped or defined properly, AI may interpret them as different materials. As a result, the walls may be treated as two separate material categories, which may lead to inaccurate analysis.

3. Unified ClassificationSystems

Just like there is a history of classifying humans, plants, and animals based on their traits, this works exactly the same way. However, this classification is done using codes to help AI detect specific elements with a standardised coding system. That is why the Unified Classification System is the most trusted one in the industry compared to the conventional Common Arrangement of Works Sections (CAWS), with a percentage comparison of 58.9% to 14.6%, respectively. For instance, the architect calls the main door Main Entrance Door and the engineer refers to it as External Steel Door. They both are right in their own context, but it is confusing for AI to understand. This is where code classification sorts things out by assigning the Uniclass code = Pr_30_59_24 (Door systems example).

4. Metadata

The “data about data” is called metadata. Such data provides access to embedded information like material origin, composition, and carbon emissions beyond its geometric attributes. It is then utilised by AI to perform intricate tasks such as evaluating embodied carbon, assessing environmental impacts over the product lifecycle, and verifying compliance with rules such as Digital Product Passports (DPPs). When elements from the BIM model include sophisticated metadata such as Environmental Product Declarations (EPDs) and material traceability, AI can inform stakeholders about the optimal environmental choice.

Constraints for AI

“Accuracy is the issue, the majority of the AI Hype Bubble at the moment is surrounding LLMs which hallucinate and generate inaccuracy but in very credible and believable ways (until someone checks the fine detail). This has to be resolved first before any AI really becomes of use within the Construction Industry where accuracy is an absolute requirement.”

~ Technical Product Specialist

1. Poor Standards

As we already know, logic is the key principle on which AI operates. Therefore, it is very important not to break that principle, which is only achievable if we use proper standards such as BS EN ISO 19650 in BIM model creation; if it is neglected, AI will consequently struggle.

Simply put, the fact is that if teams ignore standard practices, mix different modelling methodologies, or do not involve an active Common Data Environment (CDE), then results would be unreliable. Ultimately, this problem will eventually fall on humans for resolution, which is not our intended approach.

2. Inconsistent Naming

If, in a BIM model, there are variations in the naming conventions, AI will encounter conflicts while analysing the data. Humans still have familiarity with such variations. However, AI treats them as two different things.

For instance, if a modeller names a lighting fixture “Linear Light” and another calls it “Light-01”, this creates a contradiction that AI cannot deal with on its own. As a result, essential tasks such as clash detection and model checking become unreliable due to the inconsistent naming of fixtures.

3. Fragmented Information

Fragmented information means information that is scattered into pieces. It is also one of those issues AI struggles with. Due to the absence of complete data in one place, as a unified system, AI cannot see a holistic picture of the task that you want it to perform, leading to inaccurate outputs, often termed as “hallucinations”. For reliable results, AI requires everything to be on a single platform for cross-referencing and to conclude things that sound precise.

BIM Readiness Checklist for AI Integration

Before assigning a BIM model to an AI for output, ensure you review the checklist below in the model to achieve the expected results:

• Standard Alignment: Is the BIM model created using a proper standard such as BS EN ISO 19650?
• Unified Classification: Is the Uniclass classification system being implemented across all disciplines?
• Nomenclature Verification: Is consistent naming being adopted across the project?
• Parameter Mapping: Are parameters correctly mapped to the fields they belong to, in order to avoid inconsistencies?
• Manufacturer-Ready Content: Are you utilising proper manufacturer-provided BIM objects enriched with correct information?
• Metadata Completeness: Does each material have complete metadata providing information about carbon emissions and environmental impact?
• Model-Specification Integration: Is the model integrated seamlessly with the specifications written in the properties to prevent data fragmentation?
• Single Source of Truth: Is the model data and updates being stored in an active Common Data Environment (CDE) or not?
• Human Supervision: Is there a technical person overseeing the workflow and approving AI results after ensuring everything is correct?

‍