AI in Construction: Large Language Models

AI dominates today’s business headlines, with many articles talking about Tech, Finance, Software… computer centric industries. But what about Construction? We’ll explore each of these topics from a Construction perspective over a few blog posts.

Hordes of vendors line up to sell you AI-powered solutions. The first question I ask is which type of AI, because AI can mean many things:

• Machine Learning (ML)

• Large Language Model (LLM)

• Agentic AI

This post explores Large Language Models, as part of a series on AI in Construction.

LLMs are having a moment in Construction. Claude, Gemini, ChatGPT, and similar tools are showing up in proposals, execution plans, work packages, and project communication. But there's significant confusion about what they can and can't do with your Construction data.

Let's clarify: LLMs are extraordinary at language tasks. They're poor at data analysis. Understanding the difference, and how to leverage one to cover the other, helps you use them effectively.

What LLMs Actually Are

LLMs are sophisticated text prediction engines. They analyze patterns in massive text datasets, then generate new text that follows those patterns convincingly.

This makes LLMs excellent for:

• Summarizing: Create meeting minutes, executive summaries, summary of executive summaries

• Editing: Rewrite angry emails, rephrase slides for executive presentations, adapt documents from templates

• Brainstorming: Generate ideas, suggest not-so-obvious alternatives, pranks for interns / birthdays

• Drafting: Write proposals, create simple work packages, narratives of BI reports

• Reviewing: Critique a draft as the intended audience, legal, PR, IR, etc

• Coding: Create formulas (Excel), M, and DAX (Power BI), generate documentation

It goes without saying that all of these use cases require guidance and supervision from knowledgeable human beings. The more context and guidance you provide, the better the result. Always check the output carefully before sending or issuing for construction.

The Volume Limitation

Today’s LLMs are limited to 1 million input tokens – less than 1MB, or one medium-sized Excel file. Your detailed cost breakdown for a 7-to-8-figure commercial project? Too large. PO line items from Procurement? Nope. One month of timesheets on a megaproject? Fuhgeddaboudit.

With increasing volumes of data coming from a variety of source and accumulating over time, LLMs simply do not have the ability to ingest enough data to perform analytics beyond short-term, simple scenarios.

Varied Outputs

LLMs struggle with direct numerical analysis and logical reasoning. They generate text that resembles analysis rather than performing actual calculations. LLMs simply cannot process numerical data directly, much less large datasets.

I spoke with an executive who said LLMs can’t handle a P6 schedule (too much data and logic), but can summarize 1-2 page P6 PDF exports (because it’s only remixing a few words). This is a classic example of LLMs imitating data analysis.

LLMs rarely give the same answer twice - send the same prompt ten times and you will get ten different responses. They are non-deterministic, which makes them great for brainstorming, feedback, or conversational applications where creativity and variety are helpful.

Most business leaders are not looking for creativity and variety in their daily dashboards. Rather, consistency, predictability, and quick comprehension are ideal. Decisions tied to non-trivial business impacts require serious and accurate analysis. For that, LLMs, by themselves, are simply unreliable. But there is a way.

LLMs and BI Tools Working Together

The practical LLM use case in Construction data: coding assistance for Excel and BI.

Example workflow:

You need to calculate labor productivity variance by contractor and trade across projects. You know Power BI can run this for you every morning, but the DAX formula is complex.

1. You describe what you need to the LLM: "Write a DAX measure to calculate variance between actual and budgeted labor hours. The result will be filtered by date, project, contractor, and trade."

2. LLM generates the DAX code

3. You create the measure in Power BI using the LLM’s code

4. You create a matrix visual to test it on a small dataset

5. You debug any errors (there usually are some)

6. You create and format the final visual

7. You deploy your report to Power BI Service

The LLM accelerated step 2. Depending on your skillset, this saved 5-120 minutes. For DAX experts, it’s usually faster to write the code than trying prompts until finding one that’ll write the code. I also find a lot of little bugs that are easy to fix if you know DAX, but could take a lot longer to figure out if you don't. Regardless, you still own the other 6 steps. (Post Note: This is quickly changing with Agentic AI, which should also handle step 3 soon, and maybe steps 4 and 5 eventually.)

Don’t get me wrong, this is valuable, but it’s not revolutionary. LLMs are productivity tools, not replacements for BI platforms.

Think of LLMs as assistants for your BI development:

LLMs:

• Draft code to speed development

• Generate documentation

• Create narrative descriptions of visualizations and reports

Power BI:

• Retrieve, process, store, and analyze your data

• Perform calculations on a schedule and at scale

• Generate visualizations

• Deliver reliable, repeatable analysis

You need both. They serve different purposes.

Your Practical Implementation

Start using LLMs for Construction tasks where they genuinely help. Keep your expectations calibrated. LLMs won't replace your Estimating team's judgment, your Superintendent's experience, or your Safety Manager’s eyes and ears. They're assistants for specific tasks, not general intelligence (probably for many years, despite what Sam and Elon say).

Even with LLMs helping, you still need people to:

• Define which metrics matter for decision-making

• Determine how to transform raw data into those metrics

• Validate that code produces accurate results

• Interpret what the numbers mean in context

• Decide what actions to take based on analysis

LLMs make some of these steps faster. None disappear entirely.

Ready for analytics you can rely on?

Are you having success with AI on words, but struggling to do the same with project data? Need to get reliable metrics to your team on a schedule? Let's talk!

I build custom Power BI solutions that transform raw data into validated metrics to keep projects on time and on budget. Contact me today to make your data clear.