Last fall, a mid-sized civil engineering firm I work with tried something that sounded great on paper. They wanted to feed two years of project closeout reports into a large language model and make patterns appear in cost overruns, schedule reviews, and change orders. The kind of institutional knowledge that tends to disappear when a senior prime minister retires or moves on.
Six weeks passed. About $40,000 was burned. Then he killed everything.
Technology was not the problem. The data were Their closing reports were spread across three different formats, living on two SharePoint sites and a shared network drive that no one had bothered to migrate. Half of the PDF files were scanned images without any searchable text. Naming conventions had changed twice in 18 months, so the model couldn’t tell a $2 million freeway interchange from a $200,000 drainage repair. Both appeared as “Infrastructure – Miscellaneous”.
I have been consulting on AI and data strategy for AEC and manufacturing companies for about five years. This story is not unusual. I’d say some version of this plays out in almost every business that tries to jump straight to the exciting part without doing the boring work first.
I saw something similar in a regional MEP contractor, about 200 people. They wanted AI to speed up submission reviews, comparing incoming submissions to project specifications and flagging discrepancies. Solid use case, but its specs came from a dozen different architects who formatted things differently. The submissions lived in Procore, in emailed PDFs and, in one case, as photos of a whiteboard someone took on their phone. The AI suffocated and honestly so did the team trying to make it work.
Harder than it looks
What is driving this? The pitch of the AI vendors sounds amazing. Enter your documents, ask questions in plain English and get answers instantly. After years of digging through Procore exports and buried email threads, this promise looks like it might fix something. I understand why people buy.
However, what the demo doesn’t show you is the gap between the clean sample data and your actual files. In the demo, everything is formatted, digitized, consistent. In your company, data is anything your team has produced under deadline pressure over the last 10 years. It’s a very different animal.
Looking for quick answers on construction and engineering topics?
Try Ask ENR, our new intelligent AI search tool.
Ask ENR →
I’ve come to call this the “foundation problem.” It is the main reason why AI pilots stop in this industry. Everyone focuses on the technology layer. Hardly anyone focuses on the data layer underneath. Sellers won’t talk about this, because their sales process depends on whether you believe the tool works right away. Some of them think so. They have never tried running it against the file structure of a real AEC company.
Think about what a typical contractor is sitting on. PDF specs, many of them scanned from 15 or 20 year old paper originals. RFIs that live in email threads, in Procore or some other cloud platform, sometimes still in paper records in a filing cabinet. Deliveries are tracked in spreadsheets where each project manager uses a different format. Documents lessons learned that theoretically exist but no one can find. Drawings spread across Autodesk Construction Cloud, Bluebeam sessions, and local folders that were supposed to sync but didn’t.
Asking an AI to make sense of it is like handing someone a shoebox full of receipts from five different accounting systems and telling them to do their taxes. The tool is not broken. You just gave him garbage to work with.
And companies are spending real money on it right now. A 2024 survey by the Royal Institution of Chartered Surveyors showed that builders highly value the potential of AI, but the actual implementation on the job sites still lags behind expectations. A pattern is forming: enthusiastic pilot, silent cancellation, no one talks about it.
Solving the foundation problem
The companies I’ve seen really get value out of AI are doing something a lot less sexy. They are cleaning your data. Working with a checklist no one wants to talk about before approaching an AI tool.
They start by choosing a project type and centralizing documents. not all Not the last five years of each project. Just one type. Freeway bridges, K-12 schools, water treatment plants, you name it. Get all relevant files in one place with a naming convention that makes sense.
Then they cover the topic of optical character recognition. If your specs and submissions are scanned PDFs, and in my experience that’s 30-50% of what 20+ companies are sitting on, then AI can’t read them. final point Running OCR on old documents is a tedious and thankless job. Skip it and your AI will be effectively blind to half of your institutional knowledge.
After that, they find the task that eats up the most hours. Not the flashiest use case. Not that the CEO raved at a conference last month. The task in which an engineer spends more than four hours a week basically just looking for information. Specifications searches. Cross-references in the presentation. Security reports. It becomes the first goal of AI, because the return on investment is obvious and you can measure it.
The step that most companies skip? Find someone within the organization who truly feels the pain. AI pilots forced from the top, with no real buy-in from any project manager or lead engineer, stalls every time. The champion does not need to be a technician. They just need to be frustrated enough with the current process to spend a couple of hours a week trying the new one.
How to prepare for AI the right way
I saw one company doing well, a structural engineering group, about 80 people. They spent 10 weeks doing nothing but organizing their steel connection design library into a consistent, searchable format. No AI involved. Just a junior engineer with a spreadsheet and clear naming rules. When they finally connected this cleaned dataset to an AI tool, engineers who used to spend 45 minutes searching for a comparable connection detail were getting answers in less than two minutes. It worked because the data was ready for it.
No one is going to write a press release about a 10-week filename project. But this company has a tool that works. The other company received a bill for $40,000 and a folder full of meeting notes explaining why the pilot failed to deliver.
There’s a security angle here as well, and I don’t think the industry pays enough attention to that. I’ve seen engineers from ITAR-regulated aerospace suppliers paste their own part specifications directly into ChatGPT because they needed a quick summary and the internal tools were too slow. They are not being careless. They are having resources. However, compliance exposure is real. Data entered into public AI tools is by no means guaranteed to remain private, and for companies handling ITAR-controlled technical data or operating under strict NDAs, a wrong indication could be a violation. This happens to clueless companies, because no one thought to ask what their engineers are pasting into their browser windows at 10 o’clock on a Thursday night.
Do the hard work first
The answer is not to ban these tools or pretend the technology doesn’t work. It works Sometimes very well, but only when the conditions are right. The real move is to get the database in order first, create a secure internal environment where the AI can run on your actual project data without sending it to a third-party server, and scale from there.
Companies that figure this out over the next two years will have a real advantage. Not because they chose better software. Because they did the unglamorous job of making their institutional knowledge accessible and structured. Companies that continue to chase demos will continue to write off pilot costs and wonder why this technology that everyone seems excited about isn’t working for them.
Construction has always been better at building things than at organizing information about what it builds. AI does not solve this problem. It just makes it impossible to ignore.
Alex Ryan is CEO of Ryshe, an AI and data consulting firm that serves as the data and AI arm of Wiley|Wilson, a 125-year-old architecture and engineering firm. He works with AEC, manufacturing and aerospace companies on data strategy and secure AI implementation.
