An open review of strengths and weaknesses of Construction AI (and in the AEC industry

When trialling AI solutions available to the construction industry you quickly discover both the unbelievable potential of the technology alongside the apparent limitations of the technology as we see it today, as far as construction use cases are concerned.

In the construction industry we care about accuracy, privacy and confidentiality, efficiency and dealing with many uncertainties. In this article I explore how the technology currently sits against these criteria. This is based mainly on my experiences using the Construction AI software named

Strengths of AI in Construction:

1. Endless automation of repetitive tasks: AI excels in automating routine tasks, from filling out reports and forms to employing image recognition for defect detection. Unlike humans, AI systems don’t experience fatigue. To determine if AI can address your challenge, consider if the task is a repetitive process governed by specific rules or approvals needed for progression. This typically involves tasks that can be completed through reading or visual inspection. If your task fits this description, it’s likely an AI-solvable issue. The complexity of the rules and frequency of the task should be assessed to ascertain the cost-effectiveness of AI automation.

2. Deep options analysis: Early chess computers, exemplified by “Deep Blue” created in 1995, demonstrated the superior capability of computers to search and analyze far more possibilities than humans. In the realm of chess, for instance, a novice human player might assess 1-2 moves ahead, while a human Grandmaster could project 5-8 moves. In contrast, contemporary chess engines like Stockfish can examine 20-30 moves ahead in certain scenarios. This principle can be applied to AI in construction, enabling these systems to evaluate potential outcomes of a construction project more effectively than humans. Additionally, AI can access and retrieve information from project archives with greater efficiency and thoroughness than human searches.

3. Grappling staff shortages: The construction industry is grappling with a skills shortage, with numerous companies reporting unfilled positions for over a year. Automation offers a solution to enhance productivity despite limited resources. This issue is compounded as many countries struggle to manage aging infrastructure, further straining resources and necessitating more efficient work practices. AI-driven automation can assist by generating initial drafts, composing meeting minutes and summaries, sourcing information, and thereby freeing up time for more high-value tasks.

4. Organising disorganised information: The construction sector often deals with extensive archives of PDFs and disorganized folder systems containing multiple versions of the same document. AI technology can streamline this process by introducing order through techniques such as semantic similarity, a method utilized by This approach enables efficient searches across thousands of construction documents, allowing for the rapid retrieval of information and the instant answering of construction-related queries.

Limitations of AI in Construction:

1. Difficulties in communicating uncertainty:

Large Language Models (LLMs) have particularly struggled with conveying uncertainty. This issue is accentuated because AI developers aim to deliver concise answers to minimize operational costs, and users prefer succinct responses for time efficiency and quicker access to information.

2. Over-expectation: Closely related to the initial challenge, AI applications often require a period of fine-tuning for specific use cases and are seldom ready-to-use solutions. Consequently, it becomes crucial to manage user expectations and establish clear milestones using smaller samples or datasets before full deployment on a project.

3. Specialist applications: In the construction industry, numerous subcontractors or specialists are employed to perform specific tasks. Due to the high costs of training AI models and the scarcity of publicly available data for AI training, users of AI applications in specialized fields often find the results less impressive than expected.

4. Barriers to entry: A significant number of companies lack internal data science teams to develop and train their own models, and recruiting skilled data scientists is both expensive and challenging. As a result, in-house developed prototypes often fail to meet expectations.

5. Privacy concerns: In many construction projects, the use of sensitive company data necessitates extra precautions. This includes running AI applications on the construction company’s own cloud environment to prevent data transmission to external third parties, or anonymizing any data that is sent externally, ensuring enhanced privacy and security.

6. Silo’d data: All AI applications depend on data, and the construction industry has an abundance of it, though it’s often isolated or embedded within PDF documents. Addressing this challenge, we at have dedicated the past two years to creating tools designed to unlock and make accessible the wealth of data contained in your construction documents. in particular is a standout amongst the new wave of construction automation startups sweeping across construction and seeing that they are working with the likes of AECOM, ARUP and Mott Macdonald is no surprise as early adopters begin piloting and experimenting with this next generation of automation.