Where AI actually works in construction operations (and where it doesn't yet)

I spent 14 years in construction operations at Pulte. In sales, land development, construction management across Arizona and Southern California. I've coded invoices for six hours straight while covering three departments. I've run 5 to 20 permit requests a day through a process that involved tracking, coding, approvals, printing documentation, getting checks cut, and shipping everything to the city or county by FedEx.

There were stretches where that was my entire day. Just moving paper through systems.

I think about that now and it kind of makes me sick, because a huge chunk of it could have been automated. Some of it I actually did start automating towards the end with Power Automate, getting files saved into the right folders automatically, getting documentation printed and staged for me. That alone started giving me hours back.

So when someone asks me where AI fits in construction, I don't have to guess. I lived the workflows that AI is built to fix. And I also lived the ones where people assume AI helps and it really doesn't. Not yet.

The stuff AI handles right now

The best targets for AI in construction are the repetitive, multi-step admin workflows that eat your team's day and don't require judgment. The kind of work where the steps are always the same, the inputs are predictable, and the person doing it is basically acting as a human router.

Permit tracking is a good example. The process I described above, requesting permits, logging them on a tracker, saving them into a system, filing them on a drive in the right format, coding them, routing them through approvals, tracking the check, pulling the documentation, sending it out... that's a sequence. Every step follows the one before it. The inputs are consistent. A human doing that work isn't making creative decisions, they're executing a checklist. That's exactly where automation pays off immediately.

Invoice coding is similar. When I was covering three departments at Pulte and coding invoices for six hours a day, the work wasn't hard. It was just volume. Matching invoices to cost codes, routing them to the right approvals, tracking what went through and what's still pending. All of that can be automated now, and for most construction companies it should be.

Document management is another one. RFIs, submittals, daily reports, change orders. The filing, the version tracking, the routing. Most companies I walk into still have this scattered across email, a shared drive, and somebody's desktop. AI and automation tools can centralize that and keep it organized without anyone changing how they work. They just stop losing things.

Meeting transcription to action items is probably the fastest win I see for construction teams. You run your Ops meeting or your weekly sub coordination call, the AI transcribes it, pulls out the action items, and you've got a clean list before you leave the trailer. That one saves PMs hours every week and the adoption rate is high because it doesn't require the PM to change anything about how they run the meeting.

The stuff everyone assumes works and doesn't

Here's where I'm going to be honest in a way most AI consultants won't be, because I think it matters more than looking impressive.

AI reading and responding to your email is not there yet. I've tested this. I've set up systems that run reports and drop them in my inbox every morning, and they miss emails I know came in. The information just isn't all there. And even when it catches everything, the idea of AI responding to your emails for you in construction is a problem. Your subs, your clients, your inspectors... those are relationships. The way you respond to a frustrated trade partner is different from how you respond to an owner asking about timeline. That's judgment, not information processing. I don't think AI should be doing that for you.

AI-powered estimating and takeoffs is the other one. Every vendor has a demo that looks incredible. But when I talk to estimators who've actually tried to use these tools on real projects, the accuracy isn't reliable enough to trust. A good estimator brings years of knowledge about local conditions, trade availability, material pricing fluctuations in their specific market, and the quirks of how their company prices work. AI can help organize the data. It can speed up the math. But the experienced judgment that makes an estimate accurate... nobody's been able to replicate that yet. I think it'll get there. It's not there today.

What I see in almost every company I audit

The pattern is pretty much always the same. Spreadsheets everywhere. People doing workarounds because they've been doing it that way forever and it's just how the company runs.

I have an estimator right now who has his own templates, his own Excel formats, his own way of doing everything. He hasn't wanted to adopt any other way. And the thing is, his process works. For him. But everything lives in his setup, on his machine, in his format. When something disrupts that, even something small, the whole thing stalls.

I watched a project drag for three weeks because the operator's laptop broke and they had to go through IT to get another one. Three weeks. Not because anything went wrong with the project. Because everything that mattered was on one person's computer and nobody else could access it.

That's not an AI problem or a software problem. That's an organizational problem. And it's the reason I tell every construction company that asks me "where do we start with AI" the same thing.

Start with an audit, not a tool

The honest answer is you start by understanding what you actually need. Not what a vendor is selling. Not what you saw in a LinkedIn post. What your people are actually doing every day, where the time goes, and what's easy to automate versus what requires real human judgment.

That means surveying the team. Finding out where people's comfort levels with AI actually are, because the range is huge. Your PM might already be using ChatGPT for email drafts while your field crew has never opened it. Your estimator might be completely set in his ways while your office manager is ready to automate everything tomorrow.

The low-hanging fruit is always the move. Start with the boring, repetitive stuff that nobody likes doing and that doesn't require expertise. Automate that. Show people it works. Let them see the time come back. Then build from there.

I've seen companies try to roll out AI across the whole organization at once, buying Copilot for everyone and pointing them at the tutorials. It doesn't stick. The tutorials don't teach Copilot on your bid templates or your subcontractor records. People poke at it for a day and go back to the way they've always done it.

The companies that actually get AI working treat it like any other operational change. One piece at a time, starting where the pain is loudest, and making sure the people who do the work feel confident before you move to the next thing.

The real reason most AI rollouts fail in construction

It's not the technology. The tools are good enough. It's that most companies treat AI like a technology problem when it's actually an organizational one.

The estimator who has his own templates and his own way of doing everything... he's not wrong. His process works. The question is whether the company can afford to have critical workflows living in one person's head and on one person's laptop. That's a business risk question, not a technology question.

AI doesn't fix a broken process. It speeds up an existing one. So if the process lives in spreadsheets and email and one person's memory, automating it just means you've automated a mess. You have to understand the workflow first, then decide what to automate, then bring in the tools.

That's the order. And it's the order that most AI consultants and most vendors skip, because it's easier to sell a product than to do the operational work of figuring out what a company actually needs.

Where to start

If you want a rough read on where your team actually stands, the AI readiness quiz is a two-minute start. And if you'd rather just talk it through, book a free 30-minute audit and I'll tell you honestly which of your workflows I'd automate first, and which ones I'd leave alone for now.

---

Kelly Stephens is the CEO of EyeOn Automations, a custom software and AI company for construction and real estate operators. She spent 14 years in construction operations at PulteGroup before building software for the industry. Based in Southern California.