AI's Impact on Construction Workflow Automation

Raymond Levitt, Senior Advisor, Operating Partner, Blackhorn Ventures

Raymond Levitt is an Operating Partner and Senior Advisor at Blackhorn Ventures, an early-stage venture capital firm investing in startups that enhance resource efficiency and reduce GHG emissions in the built environment, transportation, supply chain and energy sectors. He co-founded and led Vité Corporation, Design Power, Inc., and Rackwise, Inc., and serves as President of the Farmers Investment Club, an angel investment group affiliated with Stanford University. Dr. Levitt has held faculty positions at MIT and Stanford University, where he founded and directed the Global Projects Center.

Recognizing Raymond Levitt’s expertise in construction technology and AI, this feature highlights his strategic approach to leveraging data and advanced technologies to transform the construction industry. By focusing on high-quality data and empowering firms to develop their AI solutions, Levitt champions integrating digital tools and automation to improve efficiency. His insights underscore the importance of innovation, data management, and adapting business models to unlock the full potential of AI in shaping the future of construction.

The Evolution of AI Technologies in the Construction Industry

There are several key differences between first-generation and modern AI systems, particularly in their sources of knowledge, transparency and infrastructure requirements. Most importantly, the first generation of artificial intelligence relied on querying human experts to understand how they made certain decisions. These insights were built into expert systems, chains of if-then rules that linked low-level data to high-level inferences. Developed at Stanford University by Edward Feigenbaum and his colleagues using the LISP programming language, these systems were quite effective. They could diagnose infectious diseases like meningitis as well as, or better than, typical human doctors. Several startups emerged from this technology, but it did not succeed in the end.

One reason was resistance. People didn’t like being told how to make decisions by computers, even if those computers had the knowledge of leading experts. Another difference was that these programs could explain how they reached a conclusion. They could say, “You said this, therefore this,” and explain their reasoning. By contrast, much of the new generation AI functions as a black box. It analyzes large amounts of data, gaining insights not from human knowledge but from pattern matching across datasets, often scraped from the internet, which contains misinformation. As a result, modern AI can be highly capable, but depending on the data used, it can also produce hallucinations or distorted output. In some cases, the results have not been good.

“If AI can achieve the same result in a fraction of the time, it reduces reliance on human hours and monetizes valuable knowledge”

A third key difference is infrastructure. Early expert systems ran on laptops or even PCs, using very little computing power. Today’s AI requires massive computing power, often provided by large data centers. So, we can summarize the differences: human knowledge vs. data patterning, explainability vs. black-box models, and low computing needs vs. large-scale infrastructure. That said, the power and capabilities of modern AI, when trained on high-quality, proprietary data, are orders of magnitude greater than early systems. These are the key differences to keep in mind.

Leveraging High-Value Data for AI Workflow Automation

In the construction industry, the most forward-looking strategic firms have been digitizing their data for some time. These include engineering, architecture, and construction firms that adopted 3D digital models early instead of relying on paper-based documents and workflows. These companies now possess valuable proprietary data. If Autodesk or another large software provider could access that information, they could generate effective AI solutions. However, strategic firms are generally unwilling to share their proprietary data, as it represents intellectual property and a significant competitive advantage. From our experience, the firms best able to automate workflows based on high-quality data are strategic, forward-looking players— architects, engineers and contractors who adopted 3D modeling early and have 10 to 15 years of digital data to train their models.

This does not mean there is no role for startups or large software incumbents like Autodesk. Their most helpful approach would be to build tools that help firms with large datasets create their own AI solutions. We can imagine platforms where third parties host their solutions, almost like an app store. Some startups are already taking this approach, building tools to integrate, organize and clean data from multiple sources. They then offer a few core apps on their platform to demonstrate what is possible and open the platform to other developers to build tools for procurement, marketing, customer support, etc. They can charge a fee for platform usage, similar to how Apple and Google run their app stores.

How Startups Can Exploit AI in the Construction Sector

Startups can build tools that help strategic players organize their data and create AI solutions. Sometimes, they can make higher-level, low-code or no-code solutions, where someone can create their procurement application. However, they often lack the data. Unless consulting firms become AI solution providers, their traditional business model of selling hours will likely become obsolete.

If AI can achieve the same result in a fraction of the time, it reduces reliance on human hours and monetizes valuable knowledge. Firms need to adapt their business models, but strategic players are the ones that can most easily succeed. Startups can assist them in this process. Startups may also gain access to data people would be willing to share, such as safety data on construction sites or data to reduce design errors. However, the most interesting startups will likely be those that build tools to help strategic players with digital data to organize, clean, and integrate it from multiple sources. Large projects generate tens of thousands of blueprints, revisions to blueprints or digital 3D models, requests for information between subcontractors, engineers, and architects, and productivity data from time cards and payroll.

If a startup can help organize all this information, it could develop an exciting solution. In some cases, startups may have enough expertise to build solutions themselves. Startups could also combine the best capabilities of first-generation and newer AI technologies by applying expert-generated rules to patterns found in data rather than simply pattern matching like some of the chat GPT-type tools.