AI in Construction: When Does the Owner See the Return?

By Terry Devlin, Founder of Devnua

For more than a year, artificial intelligence has been in the hands of project teams. Copilot licenses were purchased. ChatGPT accounts were opened. Meeting assistants started generating summaries. Drafts, transcriptions, and reports began appearing in seconds rather than hours.

The industry adopted the tools. That part is done.

Now ask the question that few in the industry want to ask out loud: what has the owner received in return?

Not the consultant. Not the software vendor. The owner. The party that funds the project, retains the asset, and lives with the long-term consequences of every decision made during delivery. After a year of AI on project teams, where is the measurable return to the owner?

What the Year Actually Bought

Be honest about what happened. Individuals got faster. A project engineer drafts correspondence in minutes. A project manager gets a meeting summary before leaving the room. An estimator pulls together a comparison faster than before.

Those are real gains. They are also personal gains. They stayed with the person at the keyboard.

Faster individuals do not equal a more capable organization. When one team member uses a private set of prompts, another relies on a different tool, and a third avoids the technology entirely, the organization does not get an operating improvement. It gets variation at higher speed.

Some of that AI use was never formally approved. Project information has been entered into systems nobody vetted. Inconsistent prompts produce inconsistent outputs. AI-generated content gets accepted because it looks complete, not because it was validated. Different teams adopted incompatible tools without an organizational standard.

So the owner paid for the licenses, directly or through consultant billing rates, and received fragmentation in return.

Why the Return Never Arrived

Three reasons explain the missing return.

First, firms automated processes that were never standardized. If one project manager documents decisions one way and another documents them differently, introducing AI does not fix the inconsistency. It automates it. Technology cannot correct an operational process that was never properly established. It can only multiply whatever discipline or disorder already exists.

Second, nobody defined the return before buying the tools. ROI requires a baseline. Most organizations adopted AI without recording how long tasks took before, what documentation quality looked like before, or where professional time was being lost before. Without a baseline, every claim of improvement is anecdote.

Third, the economics stayed upstream. Vendors collected license revenue. Consultants absorbed efficiency into margin. The owner, who funded all of it, saw little change in project outcomes, fee structures, or the quality of information delivered at closeout.

This is the uncomfortable truth behind the adoption numbers. Usage is high. Return to the owner is unproven.

What a Real Return Looks Like

The return becomes visible only when it is defined in terms the owner cares about. That means measurable categories, tracked against a standard:

Recaptured professional time. Hours shifted from administrative work back to judgment, coordination, and leadership. Measured per role, per month, against a documented baseline.
Decision cycle time. The elapsed time from RFI to answered RFI, from submittal to approved submittal, from issue identified to issue resolved. Shorter cycles with the same or better quality.
A stronger project record. Complete, consistent documentation across every project. Minutes that record decisions instead of conversations. Change documentation that holds together under review.
Less rework from missed information. Fewer disputes rooted in undocumented decisions or fragmented correspondence.
Retained institutional knowledge. When experienced people leave a project or a firm, their knowledge stays in the system instead of walking out the door.

None of these require believing vendor promises. All of them can be measured. But measurement is possible only where a standard exists to measure against, which is why standardization comes before technology, not after it.

The Condition Nobody Wants to Hear

There is no shortcut around the sequence. Process first. Automation second. AI where it adds measurable value.

Before introducing automation or AI, an organization has to understand how information moves, where decisions get delayed, where responsibilities are unclear, and which activities consume professional time without requiring professional judgment. Then the standard gets established. Then technology gets integrated into the standard, with human oversight where judgment matters.

Governance has to be designed alongside the technology, not after it. Approved platforms. Defined data boundaries. Standardized methods. Human review requirements. Decision and approval authorities. Auditability. The firms that skip this step are not moving faster. They are accumulating risk at higher speed.

This is also why the value was never in the prompt. A generic AI tool can summarize an RFI. A purpose-built system, designed around the operating environment, can track response periods, recognize recurring issues, identify cost or schedule implications, route responsibilities, maintain documentation, and flag escalation conditions. The value is the operating intelligence behind the tool.

The Work Ahead

The question facing the industry is no longer whether project teams use AI. They do, and they will continue to. The question is whether anyone is designing the return.

That is what Devnua was built to do. Owner-centered intelligent systems: operational standards, workflow automation, and AI-enabled project agents designed around the organization, with the efficiency returning to the owner rather than remaining exclusively with the consultant. The systems developed during an engagement keep producing value after the engagement ends. Project knowledge becomes organizational knowledge. Processes become repeatable. Performance becomes measurable.

A year of AI adoption taught the industry what the tools can do. The next phase belongs to the firms that decide what the tools are for.

The owner has been patient. It is time to show the return.

Terry Devlin is the founder of Devnua, a boutique advisory firm helping construction organizations build intelligent operating systems through operational standardization, training, and AI integration. AIPMCM is his industry publication covering artificial intelligence, program management, and construction modernization.