What McKinsey's new AI report for AEC means if you run tendering

McKinsey’s Engineering, Construction & Building Materials practice published its latest read on AI in Architecture, Engineering and Construction (AEC) in July 2026 (“How AI is reshaping the future of the AEC industry”, McKinsey Engineering, Construction & Building Materials Practice). The part that matters for anyone who owns bid outcomes: when they map where value lands in the first 18 months, the list starts where every project starts, with winning and pricing the work. Bid/no-bid analysis, estimating, benchmarking, scope normalization, pricing scenarios.

If your CEO forwards you one AI report this year, it will probably be this one. Here is what it says, what it means for tendering, and where we would push back.

What the report says, filtered for pre-award

The report covers the full project life cycle, more than 150 workflows across 25 domains. Filtered for the pre-award phase, three things stand out.

First, the near-term value map. McKinsey groups the first wave of AI value into five workflow families, and winning and pricing the work leads the list.

Second, their example of the operating model most firms will actually run first. To illustrate what they call an agent-led, human-accountable workflow, they describe an estimating agent that assembles the bid, proposes contingency levels and surfaces commercial risks, with a proposal manager approving what goes out the door. Of all the workflows they could have picked to explain the future division of labor between people and agents, they picked bid work.

Third, the numbers. The McKinsey Global Institute estimates AI and automation could be worth roughly $126 billion a year in European construction by 2030, and that 39 percent of nonphysical work in construction is technically automatable, rising to half in architecture and engineering. Worth citing in a board paper, with one caveat: these are technical potential figures, not adoption forecasts. Nobody captures that value by buying licenses.

Three findings that matter if you own bid outcomes

Start where the experts are scarce.
McKinsey’s playbook says to prioritize workflows where performance depends on a small number of experienced people. In tendering, that is not an abstraction. Every estimate leans on the judgment of people who have seen the project type before, and most contractors can count those people on one hand. Anyone scanning the job boards can see how many firms are advertising for estimators and bid managers right now. The report treats key-person dependency as a business risk. Pre-award is where that risk is most concentrated.

Experience has to live in the workflow, not in heads.
The report’s sharpest observation is that most project workflows are

repeatable decision systems disguised as expert judgment.

The implication for bid teams: lessons from past tenders, what was priced, what was assumed, what it actually cost, only compound if they are captured where decisions are made and fed back into the next bid. McKinsey describes the loop from estimates to actual outcomes as a competitive moat. We agree, and almost nobody has built it, because bid archives record the number, not the reasoning behind it.

Governance arrives through contracts, not regulators.
The report expects practical limits on AI use to come from clients, insurers and professional codes before formal regulation does. That means records of how AI was used, what it read, and who approved what. For bid work the bar is concrete: if a number goes into a tender, you need to be able to show where it came from. Tools that cannot trace their outputs will not survive contact with a serious client’s procurement team.

Where we would push the thinking

Two places.

The report opens on site: a superintendent photographs prefabricated pipe spools that no longer fit, and agents assemble the options within hours instead of days. It is a good story, and it is an execution story. Most of the margin was locked long before the spools arrived, in what was priced, what was assumed and what was signed. The industry keeps pointing AI at the phase where problems become visible. Most of the money is decided in the phase where they become contractual. The cheapest place to fix a project is before signature. The report’s own near-term ranking says so, but its imagination keeps drifting back to the build phase.

Second, the report treats estimating as a workflow. We think it is a decision layer. Bid/no-bid, risk pricing, contingency, terms and qualifications are not steps in a pipeline. They are one connected set of commercial judgments, drawing on the same underlying understanding of the project. Automating the takeoff while leaving those decisions disconnected gives you a faster version of the same variance. The value is in connecting them.

What to do before year end

Three moves, none of which need a program office.

Pick two or three bid workflows where outcome variance is highest. For most contractors that is scope review on incomplete documents, risk pricing, and the bid/no-bid call itself. Start there, not with a company-wide AI strategy.
Start capturing decision context at the point of bidding. Not archives, not a data lake project. Record why the number was the number: the assumptions, the exclusions, the risks you priced and the ones you accepted. That is the raw material for the learning loop McKinsey describes, and it cannot be reconstructed later.
Ask the data questions before any pilot. Who owns what, whether your tenders train someone else’s model, whether you can leave with your data intact. We published the ten questions we would ask any vendor, including ourselves.

The report’s quiet message is that the tools are arriving faster than the operating models. The contractors who come out ahead will be the ones who decide, deliberately, where judgment sits and what gets recorded around it. Tendering is the best place to start, and estimating now sits on McKinsey’s own shortlist of where to begin.

If you would rather see this on a live tender than in a report, we will run a short analysis of your actual documents, or just show you the Volve platform.

See an overview of Volve features here.