When AI Meets the Reality of Construction
AI is moving rapidly into construction, but turning its potential into practical results is where the real challenge begins.
Rhys Craigie is Head of AI and Innovation at Acuite Construction, with extensive experience across construction technology, data and emerging technologies.
Before joining Acuite, Rhys was Group IT Manager for Data and Emerging Technologies at McConnell Dowell, where he worked at the intersection of technology, data and real world construction projects.
At the Construction Technology Leaders Summit, Rhys will present AI Beyond the Tool: Integrating Intelligence Across Design and Delivery, looking at what it takes to move AI from individual tools into the way construction projects are actually delivered.
We asked Rhys three questions to give you an insight into what it really takes to make AI work on construction projects.
There is no shortage of AI tools available to the construction industry. What do you think organisations are missing when it comes to turning these tools into genuine business value?
RHYS CRAIGIE:
Organisations aren’t short of AI tools, and honestly, they’re not short of ideas either. “AI could probably do this faster” is something I hear constantly.
What’s actually missing is on the human side. There’s real nervousness about what AI means for people’s jobs, distrust and disbelief. The industry also did some of this damage to itself early. The initial message was “give it anything and it’ll give you gold.” When people fed it messy data and got messy results back, confidence was gone before most had a fair chance to build any, bar the ones who persevered.
Even where the rollout and training are genuinely good, uptake spikes and then settles because people are creatures of habit.
Part of that comes back to specificity. We’re seeing genuine value now at an individual level, with people using tools like MCP to search and interrogate huge volumes of project data in a fraction of the time it used to take. But that only works with well structured data behind it and some discipline in how the question gets asked.
Vague prompt, vague answer, and someone quietly decides, “this doesn’t work.”
A practical prompt playbook, with proven prompts for common construction scenarios, essentially AI on demand, is one of the better ways I’ve seen to close that gap while people build the confidence to ask better questions on their own.
Look past the people side, and the pattern that stands out to me is where the value is actually landing: at the individual, personal productivity level, not across the whole enterprise yet.
Good use cases are everywhere at that level. They stall at “how do we actually do this?” because most people don’t have the time, or the expertise within arm’s reach, to take an idea further.
At the enterprise level it’s a different problem again. The data, the domain expertise and the context aren’t there, the scrutiny is sharper, and governance has to hold. AI should only ever be able to surface what someone’s already authorised to see, never a shortcut around normal access controls.
When a pilot doesn’t pass that sniff test, it gets stopped, rightly. The businesses that treat that as proof AI doesn’t work, rather than proof the data and process underneath it aren’t ready, are the ones missing the lesson.
The ones getting it right share one thing. They’ve built AI into an existing process instead of bolting it on as another standalone tool. Once that’s true, good practice tends to spread through the business on its own because people start talking to each other about what’s actually working.
Your work has involved data, emerging technologies and digital delivery across major construction environments. What foundations need to be in place, particularly around data and governance, before AI can be effectively integrated into project workflows?
RHYS CRAIGIE:
The foundation I’ve learned the hard way is ownership. Who owns the data, who controls it, and who’s actually accountable for it. Making one person fully accountable for every decision within their functional area of data closes the loop. It’s what turns “how we handle this” into an approved, universal business practice rather than a guess.
The default I keep seeing instead is “that’s IT’s job”, treating data integrity, and every exception that comes with it, as something to be owned centrally by a team that was never set up to carry that decision making weight, who aren’t even subject matter experts. Without one accountable person per data domain, nobody actually owns the rules, and that’s where it falls apart.
The other half of it is exceptions. My old team had a saying about them I won’t repeat verbatim, but the gist was: there’s no room for them. If a rule doesn’t hold across the whole enterprise, and needs tailoring for every instance, it shouldn’t exist from a data perspective, because those nuances are exactly what wrecks consistency, in your BI, and in any AI response feeding a major workflow.
An AI can only work with what’s written down, and the honest answer is most of those exceptions never make it into a formal procedure anywhere. They live in someone’s head as “just how we do it.” Unless that becomes an explicit, hard coded rule as part of the transformation, the AI has no way of knowing it exists, or more importantly, why.
It’s a simple level of governance, but I’d argue it’s the actual foundation of good governance overall. Data has matured from a by-product into the product, and it deserves a seat at the adult’s table like any other business asset. That gap between raw and reliable is something I’ve seen play out directly on a project we’re running with Acuite right now.
Pulling data straight out of the underlying database produced AI insights that were confidently wrong, because the raw records were never actually what people relied on. Before anyone saw that data, it had already passed through custom fields, business specific formulas, reporting period cut offs, and a formal review and approval process.
Financial data is the worst offender for this. Timing, definitions and reporting context can completely change what a number means. Raw data is just the building material. What AI should be working from is the finished structure, the version the business has already reviewed, approved and is prepared to make a decision from.
Skip that layer and you don’t get an obviously bad answer. You get a technically plausible one that the business itself would never sign off on, which is arguably worse.
Looking ahead, how do you see AI changing the relationship between people, data and systems across design and delivery, and what should construction organisations be doing now to prepare?
RHYS CRAIGIE:
I’ve said it already, but data has to be treated as a critical, actively leveraged asset. AI even at a basic level is genuinely useful for that: “I know someone sent me this, find it.” That’s real value, but it’s still surface level, nowhere near the full potential once you’ve got highly contextualised data, good data practice, and real data maturity and awareness running through the business.
Outside construction, plenty of industries already operate that way. Inside it, in this country, I don’t think we’re there yet. A lot of businesses have only just matured to the point of having a CIO at the table, with most technology leaders still reporting up through the CFO.
Add to that, a dedicated CDO or CDAO is rare to non existent. I don’t think data will get the gravitas it needs until someone is sitting at that table advocating for it directly.
Construction still has one foot in “this is how it’s always been done”, and that needs to shift quickly. Otherwise, I think there’s a real chance newer entrants without that legacy baggage build the advantage and take ground that established players didn’t move fast enough to defend.
The people question is the more interesting one to me, and it’s one I don’t think gets enough attention. Human connection has been quietly eroding for over a decade already. Look at what tools like Teams did to actual teams, or what social apps did to how willing people are to just talk to each other.
I think AI will, if anything, accelerate that. In the near term, it can widen the gap further. A chatbot can piece together more about the business you work for than possibly the person sitting next to you can.
But I think it cuts the other way too. It makes real human relationships more critical, not less. The check work, the experts who catch what the model missed, good old human in the loop, that becomes hyper critical, not optional.
And underneath all of that is a near term risk I think we’re underestimating. If newer people lean on AI instead of the slower process of learning from someone more experienced, we risk skipping a generation’s worth of hard won experience. That’s a much harder thing to rebuild than a dataset.
Practically, that shift is already showing up as a widening skills split. Comfort with technology, data and AI is fast becoming a baseline expectation for newer construction leadership roles, but the practical judgement, critical thinking and problem solving that only comes from experience is becoming the premium skill, not a given.
For organisations, that means two things happening at once: upskilling the project managers, commercial managers and site supervisors already in the business so they can actually work with these tools, and investing in a dedicated capability to make sure the platforms, systems and data practices behind them are applied consistently, not stitched together project by project, which is how most of construction still runs.
Most importantly, the tools need to benefit the people on the ground, doing the work. There’s a talent angle to this too. People are going to increasingly expect to work somewhere that uses technology to cut the admin grind rather than add to it, and the businesses that don’t get there will find it harder to attract and keep good people, on top of everything else they’re already behind on.
Join Rhys Craigie at the Construction Technology Leaders Summit 2026 as he explores how AI can be integrated across design and delivery. View the brochure to discover more about his session and the full two day agenda.
