Practical AI in AEC: How to Start, What to Measure, and What to Avoid

In this article, we’ll cover how to build a company culture that embraces AI, where and how to start integrating it into your workflows, common reasons AI projects fail, and the governance practices that can help position your firm at the forefront of this emerging technology.

Introduction: Thinking About AI

The hype and frequent updates around AI can be daunting when trying to understand where to start. The good news is that you don’t need to jump on every shiny, new tool that launches. The key is to stay grounded: understand how AI and automation can actually help your business.

Let’s clarify the difference between AI and automation and how they offer value:

• Automation = rule-based, deterministic, great for repetitive tasks - Example: Automatically filling out a timesheet based on calendar and jobsite data

• AI = probabilistic, adaptive, pattern-recognizing systems - Example: Predicting delays in a project schedule

So why does AI matter for your firm?

• Do more with less – scale your team’s output without adding headcount or overhead
• Work smarter – process massive amounts of data quickly to uncover insights you’d otherwise miss
• Boost capabilities – assist, augment, and amplify your team’s expertise so they can focus on higher-value work
• Run around the clock – AI doesn’t need breaks and can work 24/7 (with the right cost controls in place)

Why Now?

Create a Successful AI Culture

• Dedicate 10–15% of time to experiment with AI tools and coding
• Run internal hackathons or learning cohorts
• Reassure people: AI is here to enhance, not replace
• Communicate how AI benefits their work (e.g., more time to focus on enjoyable tasks)

• Use platforms like Coursera or DeepLearning.AI
• Bring in AEC-focused consultants like YegaTech or AI in AEC for workshops
• Attend conferences like AEC Innovate or BuiltWorlds

How to Begin: A Problem-Based Approach

• Which tasks in your workflow are highly repetitive or consume a lot of time?
• Are there bottlenecks where additional hiring is being considered to meet demand?
• Where do errors commonly occur, causing rework or delays?

• The process is rule-based and predictable
• Data inputs and outputs are well-defined
• You want to reduce manual repetition

• You have quality data to learn from
• Patterns or relationships are too complex for fixed rules
• You can clearly define what a “good” result looks like
  
  

How to Start Integrating

• Allowing humans to review and correct AI outputs
• Preventing costly errors in complex or ambiguous scenarios
• Building trust and increasing adoption among your team
  
  

• Identify easy wins (e.g., AI meeting notetakers, LLMs like ChatGPT for proposal drafting)
• Spend 1–2 weeks testing and evaluating multiple tools
• Use resources like aec + tech, which has a vetted database of 500+ AEC-focused tools organized by workflow, role, and use case, to find solutions that fit your needs
• Decide whether to start with an off-the-shelf solution or invest in a custom build. Use these criteria:
• Buy when you need speed, minimal maintenance, and a proven tool exists.
• Build when you have unique needs, internal dev/ML staff, or require full control over data and functionality.

Why AI Projects Fail

• Data Problems (Quality, Readiness, and Silos): Over 80% of the time consumed in AI projects is data engineering related: collecting, cleaning, and organizing information. Poor quality or siloed data (files scattered across teams, departments or offices) makes AI far less effective. Integrated pipelines improve accuracy, reduce redundancy, and unlock better insights.
  
  
• Proof of Concept Trap: Demos are impressive, but moving to real-world value means testing with actual project data and workflows.
  
  
• Lack of ROI Focus: Not every problem is worth solving with AI. Prioritize use cases with measurable business benefits.
  
  
• Falling for Vendor Hype: Beware of tools that overpromise and underdeliver. Ask hard questions about what the tool can deliver now, not just in a future release.
  
  
• Mismatch with Real Conditions: Models that work well in isolated or controlled environments may fail under real project conditions without ongoing monitoring and iteration.

Responsible AI: Ethics, Privacy, and Governance

• Data classification: Identify whether files contain confidential client information, IP, or regulated data.
  Privacy settings: Ensure the tool is not training on your inputs (e.g., use ChatGPT Team or Enterprise rather than the free version).
• Storage location: Verify where the data is stored (US, EU, on-premises) and whether that meets contract or regulatory requirements.
• Access control: Limit AI tool access to authorized team members during pilots.
• Anonymization: Remove project names, addresses, or identifying details when possible.
• Retention policy: Know how long the vendor stores uploaded data and how to delete it.
• Audit trail: Keep a log of AI-generated outputs, decisions, and human reviews for accountability.

Frameworks to Explore

Final Thoughts

• Start where it matters: your firm’s biggest bottlenecks
• Empower your team to explore
• Use your IT budget effectively
• Stay curious and keep learning

References