A figure circulating on LinkedIn credited a large GC with saving 70,000 man hours through AI. On the Bricks & Bytes round table, Dustin DeVan pulled it apart, and the logic travels well beyond one number. A general contractor bills the client by the people on the project. Make a project engineer 25% faster and they stay on the job for the full duration anyway. Nothing moves on the P&L. MIT’s research found 95% of generative AI pilots produce no measurable profit impact, and construction has its own version of that gap.
Dustin had been sitting on this one since he saw the post at an airport. Somebody had cited a stat crediting Turner with saving 70,000 man hours through AI. He was careful about the sourcing and said openly that he did not know whether Turner had actually said it. What bothered him was not the provenance. It was the arithmetic underneath.
His question was simple enough to be uncomfortable. The only way you genuinely save 70,000 man hours is by letting people go or not hiring them. He had seen no report of either. So did it make the company more profitable? And would a GC that size realistically cut headcount because of it?
Martin relayed a comment from the live audience doing the division: 70,000 hours across a 2,000-hour working year is about 35 people, probably absorbed through early retirement. Patric stepped in to say that missed the point entirely. The number being plausible was never in dispute. Whether the efficiency shows up on a balance sheet is.
The billing model is the thing nobody accounts for
A GC’s revenue is a function of staffing, so efficiency has nowhere to land
Here is the mechanism Dustin laid out, and it is worth sitting with because it applies to almost every professional services relationship in construction. A general contractor bills the client based on the people assigned to the project. That is the commercial structure. So when a tool makes a project engineer meaningfully faster, the engineer does not leave the project. They are staffed for the duration.
Which means the financial structure of the project has not changed at all. Dustin went further and said it probably costs slightly more, because now you are paying for the agent as well as the person. Owen’s version was that you have added resource rather than removed it. Somebody now has to manage the AI.
Dustin was not arguing that the gains are fake. His position was that contractors are genuinely trying to build the best product and properly staff the work, and they will not pull a project engineer off a job until they are confident they can still deliver. If the market gets competitive enough that projects can be run with fewer people, the market will adjust. That has just not happened yet.
People conflate time savings with cost savings.Dustin DeVan, founder of Ediphi, on the Bricks & Bytes round table
The twenty-year test
Construction tech has had its boom, and project teams have not shrunk
The line that should be printed on a wall somewhere in every ConTech office: at no point in all of construction technology have we shown that we manage a project with fewer people than we did 20 years ago. And we have had the whole tech boom.
Dustin was fair about the caveats. There are real savings in places. Slightly higher quality, a bit less rework, companies marginally more profitable. What he disputed is that any of it is material enough to change what a building costs. If a technology saves someone two hours and does not change the total money a business has to spend, that is a better working day rather than a business case.
The wider data backs the shape of this. MIT’s NANDA research found 95% of generative AI pilots delivered no measurable profit and loss impact. Morgan Stanley found only 21% of S&P 500 companies could cite a measurable AI benefit at all. BCG’s 2026 survey of 1,800 executives put the share generating meaningful financial value at 26%. Task-level gains are real and well documented, somewhere between 14% and 55% depending on the study. They just keep dying on the way from one person’s workflow to the company’s output.
Construction’s own numbers sit in the same place. Bluebeam found 27% of AEC professionals using AI in operations. Dodge and CMiC found 87% of contractors expect AI to transform the industry while 19% have actually adapted workflows for it. We covered what that gap looks like when firms try to close it in KP Reddy Co. Launches Embedded AI Transformation Practice for AEC Firms.
The construction briefing that checks the numbers first
No vendor press releases. No hype cycle. Just sharp takes on the data and decisions reshaping construction.
Join 3000+ ReadersNobody in the room could answer the measurement question
A London round table of large contractors, and not one workable method
Owen brought a data point from outside the podcast that made the abstract concrete. At a Bricks & Bytes round table in London a few weeks earlier, with people from large construction companies in the room, a good chunk of them raised the same question unprompted. How are we measuring productivity gains from AI? Nobody had an answer. Nobody had seen the impact yet either.
That is the real story. Not that AI does not work. That most organizations have no baseline, no defined success criteria, and no method for converting a task-level improvement into a number a finance director recognizes. Researchers studying failed pilots landed in the same place from the other direction: most pilots launch without predefined success criteria, so there is no way to declare success even when the technology performs exactly as designed.
Patric took it a layer deeper. His argument is that in a lot of corporates, AI simply makes people less busy. The efficiency is genuine at the desk and invisible in the accounts. Dustin’s addition, delivered with the bluntness the show is known for, was that plenty of corporate roles were not fully occupied to begin with.
What to measure instead
Four tests that turn an AI claim into a business case
- Does it change the staffing plan? If the same people are on the project for the same duration, the cost base has not moved. Hours saved is an input metric.
- Does it change what you bill or what you bid? A genuine efficiency should show up in a more competitive number, or in margin on the same number. If neither, it is a quality-of-life improvement.
- Does it reduce rework rather than reporting? Rework is where the money actually leaks. Faster reporting on the same rework rate is motion, not progress.
- Did you have a baseline before you started? Without a documented pre-AI measurement on the specific workflow, you cannot prove anything either way, which is how most pilots end up in limbo.
The pattern here matches something we found in the Revizto data earlier this year: the money disappears in the spaces between teams and between tools, not inside any single task. Our breakdown of that is in Construction’s Coordination Problem Is Now a Budget Problem. Automating a handoff is a smaller and less exciting claim than saving 70,000 hours. It is also the one that tends to survive contact with a P&L.
| Claim type | What it usually means | Does it hit the P&L? | How to verify it |
|---|---|---|---|
| Hours saved per task | A person completes a defined task faster | Only if headcount or duration changes | Compare the staffing plan before and after on the same project type |
| Aggregate man hours saved | Task-level gain multiplied across a workforce | Rarely, unless roles were removed or not backfilled | Look for corresponding hiring or headcount changes in the same period |
| Rework reduction | Fewer clashes, fewer change orders, fewer redos | Yes, directly | Track rework as a percentage of project cost against a documented baseline |
| Faster reporting or documentation | Administrative output produced quicker | Usually not on its own | Check whether it changed a decision, a bid, or a schedule outcome |
| Bid win rate or pricing optionality | Better decisions earlier in preconstruction | Yes, and it compounds | Measure hit rate and margin on bids using the new process versus the old |
We cannot confirm it, and neither could the person who raised it. Dustin DeVan described seeing the figure in a LinkedIn post that credited Turner, and said explicitly on the show that he did not know whether Turner had said it. Bricks & Bytes has not been able to trace the number to a primary Turner source, so treat the attribution as unverified. The argument in this piece does not depend on the specific number or the specific company. Patric made that point on the show, noting a figure that size is entirely plausible for an organization of Turner’s scale.
Because of how the work is priced. A general contractor’s fee is tied to the people it assigns to a project, and those people stay for the project duration. Making an individual faster does not remove them from the job, so the project costs the client the same amount. Dustin’s view is that contractors will not reduce project staffing until they are confident they can still deliver, and that competitive pressure would have to force the shift. He also noted the cost can tick up slightly, since the software itself is a new line item on top of the existing team.
MIT’s NANDA report found that 95% of generative AI pilots delivered no measurable profit and loss impact. S&P Global reported 42% of companies abandoned most AI initiatives in 2025, up from 17% in 2024. BCG’s 2026 survey of 1,800 executives found only 26% had generated meaningful financial value. Morgan Stanley found just 21% of S&P 500 companies could cite a measurable AI benefit. Task-level gains, by contrast, are well evidenced at between 14% and 55% depending on the study and task. The gap is organizational rather than technical. (Source)
The wins showing up in the data are unglamorous and specific: issue tracking, clash detection workflows, and reporting handoffs. Revizto’s 2026 survey of more than 2,000 AEC practitioners found the firms getting returns from automation had identified narrow, repetitive tasks that consume coordination time and automated those, rather than attempting to automate complex judgment calls. Coordination failures have climbed to the third biggest driver of rework, which makes handoff automation a rework play rather than a speed play. (Source)
Martin argued on the show that they might, on the grounds that owners have historically depended on contractors holding the information and on assembling a panel of specialist advisers to check the work. Owen’s read was that owners now start from a better position than they did a year or two ago, though whether that changes outcomes is unproven. Dustin separated preconstruction from execution, arguing owners are getting more involved in understanding pricing optionality because escalation, limited supply, and delivery times now materially affect total cost. He framed it as a capital allocation problem, which is also what his company works on. (Source)
Pick one workflow rather than a department. Document the baseline before anything changes, including cost and error rate, not just time. Define in advance what success looks like as a financial outcome, and agree who owns the measurement. Research on failed pilots consistently points to the same causes: no predefined success criteria, no data owner, and scoping designed to impress a steering committee rather than solve a workflow problem. Vendor-led deployments in the MIT data succeeded roughly twice as often as internal builds, which is a strategy finding rather than a technology one. (Source)
Related Articles
Construction’s Coordination Problem Is Now a Budget Problem
KP Reddy Co. Launches Embedded AI Transformation Practice for AEC Firms
Data Centers Are Keeping US Construction Afloat
Forbes: AI Productivity’s $4 Trillion Question
AI ROI in 2026: why enterprise AI fails and what works
Revizto: Bridging the Gap 2026
Microsoft Work Trend Index 2026 analysis
The AI productivity paradox, 2026 data
