Could AI Have Prevented $3 Million in Construction Change Orders?

An examination of a $30 million institutional construction project raises an important question: how many costly changes could have been identified before construction began?

By John Burton, CEO and Co-Founder, Core Runway

A $30 million institutional construction project ultimately generated nearly $3 million in change orders.

That's approximately 10% of the original construction budget.

Some changes during construction are inevitable. Building owners revise their requirements, unexpected site conditions emerge, and projects evolve. But other changes arise from issues that were already present in the design documentation before the first contractor arrived on site.

A missing specification. A conflict between an engineering drawing and an institutional design standard. An equipment requirement that was never incorporated into the project documents.

These are precisely the problems that more systematic design review could identify earlier.

At Core Runway, we've been examining the change-order history of a completed institutional project to explore a straightforward question:

How many of these changes might have been avoided if the project's design documentation had been systematically checked against the owner's requirements before construction began?

The Problem Isn't a Lack of Documentation

Large institutional construction projects generate enormous volumes of technical documentation.

A university building might involve:

• Thousands of pages of architectural and engineering specifications.

• Hundreds of drawings covering multiple disciplines.

• Institutional design standards developed over decades.

• Building codes, regulatory requirements, and referenced technical standards.

• Thousands of product requirements and construction decisions.

• 
  

Each document may be individually correct while the overall project contains inconsistencies.

An electrical drawing might specify equipment that conflicts with an owner's engineering standards. A mechanical specification might omit a required component.

A product might satisfy a manufacturer's published requirements without satisfying the institution's particular requirements.

These discrepancies are difficult to identify because the relevant information is distributed across different documents, sometimes thousands of pages apart.

And as designs evolve, the problem gets harder.

Not All Change Orders Are Preventable

It's important to distinguish between changes that could reasonably have been identified during design and those that could not.

In our analysis, we're separating change orders into several categories:

Change-order cause	Potential for earlier detection
Conflicts with owner design standards	High
Missing specification requirements	High
Conflicting drawings and specifications	High
Missing or incorrect equipment requirements	High
Incomplete design details	Moderate to high
Unforeseen site conditions	Generally low
Owner-requested scope changes	Generally low
Market-driven price changes	Low

These are preliminary classifications, not measured prevention rates.

The distinction matters. AI shouldn't be credited with preventing changes that were unavoidable or outside the scope of the documentation available for review.

But when a change results from a requirement that was already documented and simply wasn't incorporated into the design, there's a compelling argument for identifying that issue earlier.

What Would Even a Modest Improvement Be Worth?

Consider the economics of our $30 million project.

Illustrative savings scenarios

Share of change-order costs avoided	Potential savings
10%	$300,000
20%	$600,000
30%	$900,000
40%	$1,200,000

These figures are scenarios, not demonstrated savings. Actual avoidable costs depend on the reasons for each change order, the documentation available at the time, and whether earlier detection would have eliminated or simply reduced the cost.

Nevertheless, the financial implications are significant.

Even preventing a relatively small percentage of avoidable changes could justify investment in more comprehensive design compliance checking.

And these calculations don't capture every possible benefit of early detection, such as reduced disruption, fewer clarification requests, or improvements in project scheduling.

Why Traditional Design Reviews Miss Things

Architects and engineers already invest considerable effort in quality assurance and quality control.

The issue is not a lack of expertise. It's the scale of the information that must be checked.

A human reviewer might need to connect a paragraph in a university's engineering standards to an equipment schedule, a technical specification, and a drawing detail.

That comparison becomes even more difficult when documents are revised repeatedly.

Research supports the broader importance of better information management. A 2021 study by Autodesk and FMI, involving more than 3,900 construction professionals, found that organizations with stronger data practices reported fewer delays, budget overruns, rework events, and change orders.

The challenge is making the correct requirements available to the right people at the right time.

A Different Approach: Continuous Compliance Intelligence

Core Runway is developing an AI-powered platform that examines institutional design standards alongside architectural specifications, engineering drawings, revisions, and construction submittals.

Rather than relying solely on a general-purpose AI question such as "find problems in these documents," the platform structures requirements into individually evaluated rules.

Each requirement can be compared against the relevant project documentation.

Findings are categorized to help reviewers understand what action may be necessary:

Non-compliant: A specification or drawing conflicts with an applicable requirement.

Partially compliant: The design satisfies part of a requirement but omits an important element.

Stricter compliance: The design exceeds the owner's requirement, potentially creating an opportunity for value engineering.

Missing reference: The documentation does not explicitly reference a required standard or code.

Manual verification required: The available evidence is insufficient for a reliable automated determination.

Crucially, findings are linked to their source documents so architects, engineers, and building owners can review the evidence and make the final decisions.

The goal is not to replace professional judgment. It's to help experienced professionals focus their attention on issues that might otherwise be overlooked.

The Real Opportunity Is Before Construction

A design discrepancy identified during preconstruction may require a specification correction, drawing revision, or clarification between engineering disciplines.

The same discrepancy discovered after equipment has been purchased or installed can trigger significantly more work.

That is why the timing of a finding can be as important as the finding itself.

For institutional owners operating extensive building portfolios, the opportunity extends beyond individual projects.

Once an organization's design standards have been structured into a reusable compliance system, those requirements can be evaluated consistently across multiple projects and design teams.

Over time, approved exceptions, recurring decisions, and institutional preferences can also become part of a continuously improving knowledge base.

Back to the $3 Million Question

We're not claiming that AI could have eliminated the entire $3 million in change orders on our example project.

It couldn't.

What we're investigating is how many of those changes were associated with detectable problems in the original documentation, and what the financial impact might have been if those problems had been identified earlier.

The analysis requires more than reading the change-order descriptions. Each candidate needs to be connected to its underlying design requirement, original project documents, and the reason the change became necessary.

That's the standard of evidence we believe construction AI should meet.

The question isn't whether AI can find discrepancies in construction documents. It increasingly can. The question is whether those findings can prevent enough expensive problems to materially improve construction economics.

For universities, hospitals, airports, and other institutional owners managing hundreds of millions of dollars in annual construction spending, even modest improvements could have substantial financial consequences.

And that's a question worth investigating.