McKinsey says 39% of your office work is automatable

The McKinsey number has been circulating in every construction trade publication this month. I've had three owners ask me about it in the last two weeks, which tells me it's worth addressing directly.

What happened

McKinsey Global Institute published a report in mid-July estimating that AI and automation technologies could produce roughly $228 billion in annual value for the U.S. AEC industry by 2030. The part getting the most play: AI could automate approximately 39% of nonphysical work in construction, and up to 50% in architecture and engineering.

The report draws a clear line between two types of firms. One group is using AI to genuinely redesign how work moves through the organization - estimating, RFI management, procurement, field reporting. The other group is using it to make individual people a little faster at individual tasks. McKinsey's argument is that the second group will see modest productivity gains. The first group will see something structurally different.

Separately, Carrier acquired a company called 75F, which builds AI-powered building automation systems. That's more of a trade contractor story - controls, commissioning, smart building retrofits - but it signals that the larger players are consolidating the AI-adjacent tools before most operators have figured out the basics.

And Caterpillar announced a pilot with Nvidia to put an AI assistive system inside a mid-size mini-excavator. Early days, narrow application, but worth noting that the equipment manufacturers are in this now, not just the software vendors.

Why it matters for your shop

The $228 billion figure is real but it's an industry-wide ceiling, not a per-company roadmap. I'd hold it loosely.

What I think is more useful is the two-camp framing. After 12 years in construction, I've watched this pattern play out with every technology wave - CAD, BIM, cloud PM platforms, mobile field tools. The shops that got lasting value were the ones that changed how work was handed off, not just which tool the estimator used on Tuesday.

The 39% automatable figure applies to nonphysical work: scheduling, document control, RFI and submittal workflows, reporting, contract admin, some of estimating. If you're a $10M specialty contractor with a two-person office, 39% of that work is a meaningful chunk of someone's week. If you automate it well, you don't necessarily cut headcount - you give that person capacity to do the work that actually requires judgment.

The honest caution: most shops I talk to aren't close to capturing that 39% because the underlying data is a mess. Job cost codes aren't consistent. Project files live in three places. Handoffs between field and office run on habit and tribal knowledge rather than any documented process. AI doesn't fix that. It amplifies whatever structure already exists, which means the operators who will see real returns are the ones who've done the unsexy work of cleaning up their own data and documenting how work actually moves.

The Carrier/75F and Caterpillar/Nvidia stories are real, but they're not your immediate problem. What's in front of most owner-operators right now is much simpler: meeting notes that never become action items, weekly reports that take two hours to produce and get skimmed for thirty seconds, and estimating processes that live entirely inside one person's head.

Fix those first.

One thing to try this week

Pull up one recurring internal meeting - owner meeting, weekly PM sync, whatever runs on a regular cadence. Before the next one, write down exactly what a useful output from that meeting looks like: decisions made, open items with owners, anything that needs to go to the field.

Then try running that meeting with Granola or any AI note-taker, and use the output to draft that document automatically. It takes about fifteen minutes to set up the first time. The goal isn't to save time on the meeting - it's to get consistent, usable records out of something that currently produces nothing.

That's the 39% in miniature. One workflow, made systematic. Do that ten times across your operation and you'll have something real.

• Austin

Source: https://finance.yahoo.com/technology/ai/articles/ai-could-automate-more-third-132000965.html