Bluebeam published its 2027 Building the Future: Bluebeam AEC Technology Outlook on 13 October, and the headline figure is a number that did not move. According to the release, which carries a Pasadena, California dateline, 28% of US architecture, engineering and construction firms are running AI in production, only 7% have embedded it across multiple core workflows, and 40% of those already running AI in production report AI-driven savings above $500,000.
The report puts last year’s comparable figure at 27%.
Bluebeam, part of the Nemetschek Group, says the survey behind it covers 750 US AEC technology decision-makers at manager level or above, fielded online in July 2026 by Hanover Research. The report’s own summary page adds that those 750 are the United States portion of a 1,100-respondent study.
Where the Industry Actually Sits
The report uses a five-point adoption scale, and the full distribution is more informative than the 28%:
- 7% — AI embedded across multiple core workflows
- 21% — AI supporting a few specific workflows
- 32% — piloting AI in limited or experimental use cases
- 31% — not using AI, but exploring future use
- 10% — not using AI, and no plans to
The 28% is the first two rows added together, and the report is explicit that the grouping is its own: “in production” is defined there as the combined share reporting either that AI is embedded across multiple core workflows or that it supports a few specific workflows, and the document states that the grouping is Bluebeam’s rather than a survey response option.
What grew, on these numbers, is the middle. A third of the sample is in pilots and nearly another third is looking without having started. Only 10% rule AI out altogether, which is the same shape as the report’s finding that just 2% cite a lack of trust in AI outputs as a significant barrier. Whatever is holding this industry back, on its own account, it is not scepticism that the technology works.
The Report Argues Against Its Own Headline Comparison
The 27%-to-28% line is the sort of thing a trade outlet turns into a trend, and Bluebeam’s methodology section asks readers not to. It records that the company has published three of these reports and that the geography changed each time — eight countries in the 2025 edition, five in the 2026 edition, and three in this one, reported as two separate markets. The United States and Germany, it says, are the only markets present in all three, “which means they carry any trendline this report draws.”
The question changed too. The 2026 edition asked about AI use by building lifecycle phase and reported 27% of firms using AI; this edition uses the five-point scale above and reports 28% in production. On the report’s own assessment: “The figures are close and the direction is flat, but they are not strictly comparable and we do not put them in a table together.”
Analysis, labelled as this publication’s own: that paragraph is more useful than a good deal of the material around it. A vendor-commissioned survey that publishes the reasons its own year-over-year number is soft is doing something the category mostly does not, and it is the difference between a research document and a marketing asset. The same section states the subset bases plainly — questions on AI use, benefits and barriers went to the 448 organisations currently using AI, and questions about moving past pilots to the 210 that had done so — and notes that because the barrier question was asked of every organisation using AI, those 448 include the 210 already past pilots, so any barrier figure quoted without that split describes both groups at once.
It also flags a limit in its own sample that cuts against one of its own themes. Roughly 58% of respondents are aged 35 to 44 and under 1% are 55 or older, which the report attributes to screening for manager-level technology decision-makers without age quotas. Its comment on that is worth reproducing: “In an industry whose defining story is an aging workforce, readers should weigh the Section 4 findings with that distribution in view.”
Cost Is Cited Identically by the Firms That Made It and the Ones That Didn’t
The barriers, as reported:
- 50% — high implementation costs
- 34% — lack of internal AI skills
- 22% — resistance to change among employees
- 18% — security and compliance concerns
- 15% — difficulty integrating AI into day-to-day project workflows
The finding underneath those is the one that matters. Cost is named at the same rate — 50% — by firms that scaled AI and by firms still stuck in pilots, which the report reads as evidence that cost alone does not explain who succeeded. It attributes that comparison to its own analysis of respondent-level records supplied by Hanover rather than to the published tabulations, and says differences described as significant were tested by Fisher exact test at p<0.05.
What did separate them, among firms that got AI into production, was integration with existing systems and tools (47%), ahead of cross-team collaboration (29%) and high-quality data (26%). The report’s own one-line summary of that section states it plainly: “They didn’t have better technology. They had somewhere to put it.”
That is also the strategic argument running through this newsroom’s product coverage this autumn. Volve’s orchestration layer reconciles documents that already exist in a tender pack; Trunk Tools built Cortex to connect what it reads off a contractor’s drawings to the rest of that contractor’s project record; Structured AI’s QA/QC product links the drawing back to the model it came from. On these survey numbers, the thing being sold in each case — a connection between systems a contractor has already bought — is precisely what the 28% say got them over the line.
AI Has Learned to Predict, Not to Prevent
The benefit figures, from the subset already using AI, point somewhere specific:
- 64% — greater cost predictability
- 58% — identifying risks and issues earlier in project lifecycles
- 17% — reduced rework
Cost predictability at the top and rework near the bottom is a result about where AI currently sits, not about how well it works. The report’s own reading is that AI is concentrated in estimating, design and planning and thins out closer to the jobsite — which is where rework happens. Firms that have integrated it report reduced administrative burden (55%), improved coordination across teams (53%) and fewer project delays (47%).
Chief executive Usman Shuja, who joined in September 2023 and also serves as Nemetschek Group chief division officer of the Build & Construct division, frames the gap as an operational one: “The industry’s AI challenge is no longer adoption, it’s scale.”
The Labour Shortage Is Becoming a Digital-Skills Shortage
The workforce section is where the report’s numbers get uncomfortable. 82% of organisations surveyed say labour and skills shortages are limiting their ability to deliver projects. At the same time:
- 93% prioritise digital tool proficiency during recruitment — and 77% say it is the hardest skill to find
- 69% prioritise AI knowledge and experience during hiring — and 59% say that expertise is difficult to find
- Among firms reporting an AI effect on workforce needs, 70% say AI is increasing demand for hybrid technical and digital roles and 61% say it is increasing demand for digitally skilled workers
The report’s phrasing for what this adds up to is that the industry’s labour shortage is evolving into a digital capability gap. Put less neutrally, and this is analysis rather than a finding: the remedy routinely proposed for construction’s labour problem is gated behind a second, narrower labour problem that the same firms say is harder to fill.
How It Sits Against the Other 2026 Surveys
Two other AI-adoption datasets have run through this newsroom’s coverage this year, and the three do not measure the same thing — which is the main reason to be careful with all of them.
ServiceTitan’s 2026 Commercial State of the Trades, published 24 September from a July survey of 1,020 commercial contractors, found 62% had tried AI and 15% of users could show clear ROI. McKinsey’s July work on how much of construction’s nonphysical work AI could automate put the technical ceiling at 39%; the single-digit daily-use figure set against it in that piece — 9% — came from a separate global DEWALT survey of construction professionals. Bluebeam’s 750 are technology decision-makers at manager level or above across architecture, engineering and construction; ServiceTitan’s were commercial contractors concentrated in the mechanical, electrical and plumbing trades.
Different populations, different questions, different definitions of use — so the conclusion to draw from putting them side by side is not a number, and what follows is this publication’s reading rather than any of their findings. Each of the three reports a gap of the same shape: a substantial share of firms that has engaged with AI in some form, and a much smaller share that has made it load-bearing. And where each locates the constraint, it is not in the models. It is in the systems, the data and the people a firm would need to put them to work.
What This Does Not Settle
The report measures what firms say about themselves. It does not measure delivered outcomes on projects, and the savings figure — 40% of production users reporting more than $500,000, with another 31% reporting between $100,000 and $500,000 — is flagged in the report itself as a “Self-reported estimate, base n=210,” not an audited result. Neither this report nor the two above establishes what AI returns on a given project; they establish what managers believe it has returned.
The flat headline is also the softest number in the document, by the report’s own argument. If the geography and the question both changed, 27% to 28% is not a measurement of standstill — it is two measurements that happen to be close. The finding that survives the methodology caveat is the distribution: a third in pilots, a third exploring, 7% with AI embedded across multiple core workflows. That is a sector that has mostly bought the premise and mostly not rebuilt anything around it, and the thing it says stopped it was not the AI itself.
