Maryanne Baines brings a unique vantage point to the construction sector, bridging the gap between high-level cloud architecture and the gritty reality of physical job sites. With years of experience evaluating how tech stacks perform under the pressure of real-world industrial applications, she understands that the “least digitized” industry is also the one with the most to gain. She has spent her career dissecting how different industries adopt product applications, making her an authority on why some technologies thrive in the field while others fail to move the needle. In an industry where a single mistake can mean tearing out tons of hardened concrete, her expertise in risk-averse technological adoption is more relevant than ever.
The discussion highlights the urgent need for a 3.2% annual growth rate to meet a $22 trillion global demand by 2040. We delve into the repeatability of retail construction, where 80% of processes overlap, making it a prime candidate for AI training compared to unique skyscrapers. The interview covers the impact of computer vision in reducing delays—saving companies like Intel weeks of construction time—and the financial breakdown of how automation can slash delivery costs by 30%. Finally, the conversation explores the shift from horizontal software tools to integrated, vertical AI agents that focus on project outcomes rather than isolated tasks.
How can construction firms bridge the productivity gap to meet rising global demand, and what specific steps can transform an industry traditionally resistant to change?
To hit that staggering $22 trillion output by 2040, the industry has to undergo a fundamental shift in how it views productivity. We are looking at a required compound annual growth rate of 3.2%, which is a massive hurdle for a sector where 98% of megaprojects currently finish behind schedule and over budget. The resistance to change stems from a very real, physical fear: when you are pouring concrete or hanging steel, there is no “undo” button, and that creates a culture that is naturally risk-averse. Bridging this gap requires moving away from the “optimistic schedule” and toward data-driven reality. It starts with digitizing the invisible, turning the grit and noise of a job site into structured data that can be analyzed and optimized before the first shovel even hits the ground.
Why is the retail sector, specifically companies like McDonald’s or 7-Eleven, proving to be the ideal proving ground for training AI in construction?
Retail giants are actually some of the most prolific builders in the world, and they have something most construction projects lack: extreme repeatability. If you compare two skyscrapers, you might only find a 20% overlap in their construction process, but with a McDonald’s or a 7-Eleven, about 80% of the process remains identical from one site to the next. This high level of standardization—using the same design manuals, prototypes, and approval flows—creates a perfect environment for training AI models. It is far more practical to teach an AI how to build the next retail store based on thousands of previous examples than it is to teach it to build a unique landmark like the Empire State Building. By focusing on these repeatable “agents,” we can automate the multi-step reasoning and coordination required to manage tens of thousands of interdependent tasks across a global footprint.
What role does historical data play in dismantling the “planning lies” and optimism bias that often plague massive capital projects?
The “planning lie” is a human condition where estimators apply subjective, overly optimistic ranges to their schedules, but AI doesn’t have those emotional blind spots. Companies are now aggregating massive datasets—some as large as 750,000 completed project schedules representing over $2.5 trillion in capital spend—to create machine learning models that forecast performance with chilling accuracy. Instead of a planner “hoping” a task takes two weeks, the AI looks at how that specific task actually played out across thousands of similar projects to produce a probability distribution. This AI-driven Schedule Risk Analysis provides a level of foresight that is simply unattainable by human operators. When you can see the delay coming months in advance, you move from reactive firefighting to proactive management, which is the only way to break the cycle of budget overruns.
How are emerging technologies like computer vision and IoT transforming the physical jobsite into a source of actionable intelligence?
We are finally seeing a world where the job site itself becomes an active data source, thanks to tools like 360-degree cameras mounted on hard hats and IoT sensors on cranes. For example, Buildots allows contractors to walk through a site and automatically compare physical site conditions against the BIM model, which has helped Intel avoid roughly four weeks of construction delays per fabrication facility. Similarly, tracking crane utilization—one of the most expensive assets on a site—through machine learning can reveal patterns of idle time that were previously invisible. At the Manchester Pacific Gateway project in San Diego, this type of analysis allowed a team to rebalance crew schedules and demobilize a tower crane early. When you realize that crane rentals can cost tens of thousands of dollars per month, the ability to cut even a few weeks of idle time translates into immediate, massive cost savings.
Beyond the theoretical benefits, how can firms effectively measure the financial and operational impact of implementing these AI systems?
We look at this through a “total attributable value” model that breaks down into three very concrete components. First, there is direct cost reduction; workflow automation and early risk detection can cut delivery costs by as much as 30% in many scenarios. Second, we look at error-related costs, which typically account for 10% to 15% of a retail project’s budget; by using AI to enforce guardrails and maintain a unified “source of truth,” we’ve seen those error rates plummet to just 2% or 3%. Finally, there is the value of redeployed labor, where we see about 10% of a project manager’s capacity being freed up from routine administrative tasks. Instead of chasing down paperwork or verification photos, those experts can spend their time on high-leverage engineering and coordination tasks that actually move the project forward.
What is your forecast for the future of AI integration in the construction industry over the next decade?
I expect the construction AI market to more than double in the next few years, growing from an $11 billion industry in 2025 to nearly $28 billion by 2031. We are moving away from “horizontal” software—where you have one tool for scheduling and another for field surveys—and toward “vertical” AI agents that take responsibility for the entire outcome of a project. These systems will be model-agnostic, meaning they will sit on top of the latest versions of GPT, Claude, or Gemini, getting smarter every time the underlying technology evolves. The ultimate goal isn’t just to have a faster tool; it’s to have an autonomous agent that understands the sequence of tasks, city constraints, and vendor reliability well enough to help build an entire structure from start to finish. In ten years, the “intelligence” on a job site will be just as critical as the concrete and steel, and the firms that don’t embrace this will find themselves unable to compete in a $22 trillion market.
