AI in the AEC Value Chain — Why ROI Is Still Missing

An Industry Experimenting, Not Transforming

Recent data paints a clear picture: 75% of project professionals in construction now use AI in projects, up from 15% just two years earlier. The most common application? Chatbots, document summarization, and knowledge retrieval—used by 82% of adopters. Beyond that:

• Resource allocation – 62%
• Reporting and dashboarding – 58%
• Risk analysis and forecasting – 52%
• Task and schedule automation – 48%
• Stakeholder communications – 45%

So adoption is not the issue. The issue is translation into value. Only a small fraction of these initiatives have evolved into:

• Scaled platforms
• Secure systems
• Workflow-integrated solutions

Most remain isolated improvements rather than systemic change. And that distinction matters. The real question now is: have you started reshaping your business and operating model from the top down?

Understanding the AEC Workflow

To understand why ROI is lagging, you have to look at the structure of engineering itself.

Across disciplines—civil, mechanical, electrical—the simplified workflow follows a similar pattern:

1. Requirements → 2. Concepts → 3. Design → 4. Build

Each stage has fundamentally different characteristics. And AI impacts each of them differently today. The earlier stages are crucial to win the work and set up delivery. Yet the real profit still happens in Build. Knowledge work is important and here is AI applied but the use cases described today are not a competitive advantage.

1. Requirements: High Impact, Low Complexity

This is where AI is already delivering value.

At this stage, the goal is to translate human needs into technical specifications:

• Performance
• Safety
• Cost
• Regulations

AI is particularly strong here because the problem is information-heavy:

• Summarizing standards
• Identifying conflicts
• Structuring requirements
• Supporting decision-making

The result: clear efficiency gains. But there's a limitation—requirements are not purely technical. They involve judgment, trade-offs, and sometimes politics. AI can support, but not decide. Also you don't make the big money in this stage.

2. Concepts: Expanding the Design Space

In the concept phase, engineers explore possible solutions.

AI improves this process by:

• Generating more design alternatives
• Testing early assumptions
• Exploring trade-offs faster

This is where generative design and optimization start to matter. However, we're still early with AI. AI can expand the search space—but not yet fully define system architectures independently. Human judgment remains critical in selecting viable directions—with humans still in the loop, not being led by AI yet.

3. Design: The Real Disruption Zone

This is where AI has the biggest long-term impact—and also where complexity slows adoption.

The design phase includes:

• Detailed modeling
• Simulation
• Prototyping
• Testing and iteration

AI is already improving parts of this process, and as accelerating in AI coding Agents I see huge potential in:

• Automating CAD generation
• Running large-scale simulations
• Predicting system behavior
• Detecting failures early

But here's the catch: Engineering is not just iteration—it's validation.

Every design must meet strict physical, safety, and regulatory requirements. AI can accelerate the work, but it cannot—and will not soon—reliably guarantee correctness on its own.

As a result:

• Humans remain deeply involved
• Automation is partial, not end-to-end

This limits immediate ROI, even though long-term potential is massive.

4. Build: The Physical Bottleneck

The final stage—construction and manufacturing—is where digital decisions meet physical reality and where project cost and risk ultimately materialize. This is a sector that operates on thin net margins—often 8-9%—so even small efficiency gains matter, and mistakes compound fast.

AI supports:

• Planning and scheduling
• Quality control
• Progress monitoring
• Supply chain optimization

But it does not build the asset itself. One reason, we have made construction to complex for yet systems that are powerful but can not handle the complexity. Physical processes introduce constraints that software alone cannot solve:

• Material limitations
• Regulations
• Site conditions
• Logistics complexity etc.

Even with advances in robotics, humanoids, and world models, fully autonomous construction remains a long-term prospect rather than today's reality.

The Core Problem: Fragmentation

The biggest reason ROI is missing is not technical—it's structural. AI is being applied in fragments:

• A tool for requirements
• A tool for design
• A tool for documentation

But the real value lies in integration across the entire workflow.

Today:

• Each phase is optimized locally
• But the system as a whole remains unchanged

This is why gains are incremental, not exponential.

What Needs to Change

The real transformation in AEC will not come from better tools within each phase. It will come from connecting the entire value chain—end to end. Earlier, we described the engineering value chain simplified as a sequence:

Requirements → Concepts → Design → Build

Today, AI is mostly applied inside these steps. But the real problem sits between them. So even if each step improves, the system doesn't.

So Why No ROI Yet?

Because we are in the transition phase between:

• Experimentation
• And systemic redesign

AI is already:

• Improving efficiency in isolated steps
• Reducing manual effort

But it has not yet:

• Rewired end-to-end workflows
• Eliminated bottlenecks between phases
• Reached full trust in high-stakes environments

Until that happens, ROI will remain uneven.

The Outlook

The opportunity in AEC is enormous—but it requires patience and structural change.

Short term:

• Incremental gains
• Local optimization

Medium term:

• Integrated workflows
• AI embedded in core systems

Long term:

• Fully compressed engineering cycles
• AI-driven design ecosystems

The key insight is simple: AI is not failing in AEC. It's just being applied too narrowly. The real value won't come from optimizing individual steps—but from redesigning the entire value chain around it. And that's the real shift ahead: Not adopting AI into your business, but rebuilding your business and operating model around AI.