By Geoffrey Guilly, CEO and co-founder, Aitenders
I spent more than twenty years in this industry before I built anything, and I did not need a report to tell me it was broken. I had lived it.
This summer, three arrived anyway. In July, McKinsey published its analysis of AI in engineering and construction. Around the same time, Batimatech, the Montreal non-profit that gathers this sector to talk about its future, published a white paper of its own, built on Canadian data. In September, MIT and Suffolk, the Boston-based builder, released a joint study on construction in the age of AI. The three teams worked separately. One looked at the global picture, one at Canada, one at the United States. They reached the same conclusions.
When three independent reports describe the same problem and point to the same way out, it stops being one company’s opinion. It becomes the industry’s.
Construction stopped improving, and the data is not subtle
Start with the number that should stop everyone in this industry cold.
Batimatech, drawing on Statistics Canada, found that labor productivity in construction has grown by about a fifth of a percent a year over the past twenty-five years. In services, the figure is roughly one and a half percent. Their conclusion is blunt: at equal effort, the sector has made almost no efficiency gains in a quarter of a century.
McKinsey, looking globally, found the same shape from a different angle. Between 2000 and 2022, construction productivity improved by around ten percent. Manufacturing improved by about ninety. The rest of the economy learned to do more with less. Construction stood still.
“At equal effort, construction has made almost no efficiency gains in a quarter of a century.”
The most expensive decisions are made before anyone breaks ground
So where does the waste go? The reports point to the same place, and it is not the building site.
Batimatech cites the research behind the book How Big Things Get Done: across the past century, roughly nine in ten megaprojects have exceeded their budget, their schedule, or both. Nine in ten. That is not a problem of pouring concrete. It is a problem of decisions made long before the concrete is ordered, in the bid documents, in the assumptions, in the call to bid or walk away.
McKinsey puts the same idea in forward-looking terms. It places the first wave of AI value not on the site but ahead of it: in bid and no-bid analysis, in drafting proposals, in estimating, in tracking the obligations a contract creates. The money is made or lost upstream.
“Overruns are rarely built. They are signed.”
The reason is that the industry has no memory
Why does this keep happening? Here the reports become almost interchangeable.
Batimatech describes an industry where decisions rest on professional experience and intuition, where the most senior and best-paid person in the room decides, and where the data that might have guided them sits walled off in closed systems, as if in a silo. It calls the result an efficiency bias.
McKinsey describes the same thing and names the risk: performance that depends too heavily on a small number of experienced people. When those people leave, what they knew leaves with them. MIT and Suffolk find it too: institutional knowledge is rarely codified, and data sits siloed across owners, contractors, subcontractors and regulators. The industry does not keep what it learns. Every bid starts from zero. Every quarter without a system, the gap widens.
“The most senior person decides, and when they leave, what they knew leaves with them.”
All three describe the same answer, and it is not a chatbot
The most striking agreement is on the solution. Most firms still expect the answer to be a better chatbot. All three reports point the other way.
Batimatech draws a sharp line between what it calls individual AI, the chatbots and assistants that help one person with one task, and systemic AI, intelligence built into the critical workflows of the business itself. The first, it argues, delivers modest gains. The second is where the real value lies, and reaching it means rethinking whole processes rather than bolting a tool onto the side of them.
McKinsey says the same in its own words: the leaders will be the firms that redesign entire domains, not the ones that deploy isolated use cases. Both are describing a vertical AI operating system (OS), software made for the specific, structured, high-stakes work of winning and delivering a project. Not a general assistant pointed at construction and hoping.
MIT and Suffolk went further. They named the prerequisite: a shared data infrastructure layer that standardizes outputs at every handoff. Building it, they write, is “the most consequential near-term opportunity the industry has on which to act.” Without it, no AI lever compounds. With it, the industry finally captures what it learns. That layer is what we have been building since 2019.
“A general tool used everywhere changes little. A system built into the work changes everything.”
The consensus is forming now
I should be honest about why this matters to me. We started building Aitenders in 2019, for exactly the problem these three reports describe: to read an RFP the way an experienced engineer would, to carry the commitments from the bid into delivery, and to capture what a company learns so the next team does not start from zero. We built it in the industry, with customers, without raising venture capital, because I had lived the problem and could not un-see it.
For years, that was a founder’s conviction. This summer, three independent teams, working apart, described from the outside the company we had been building from the inside.
We built this before the consensus. Now the consensus is visible. What matters next is who has already built the answer, and whether the industry chooses to use it.
“We built this before the consensus. Now the consensus is visible.”
Sources
McKinsey, How AI is reshaping the future of the AEC industry
Batimatech, Tomorrow Is Built Today (productivity figures)
MIT and Suffolk, Construction in the Age of AI