The Next Big AI Opportunity Is in Industries Nobody Writes About

A while back I watched a transportation dispatch operation work through an ordinary day. The dispatcher had a phone in one hand and a spreadsheet open on the screen. A driver called in running late. A trip had to be moved. Another driver’s schedule changed. Every change meant another call, another cell updated by hand, another note to remember who was where.

What stuck with me wasn’t that the process was messy. It was that this wasn’t a tiny operation. It was an established business with real revenue, real customers, and a level of daily complexity that would make plenty of software startups nervous. And a big part of it ran on phone calls, text messages, and a spreadsheet.

Once you notice this, you start seeing it everywhere. Entire industries with serious money moving through them still run their core operations the way they did fifteen years ago. Meanwhile, most of the conversation about AI is happening somewhere else entirely.

Where AI is actually landing

If you only followed AI through headlines, you’d assume every business has already rebuilt itself around it. The data says otherwise. The U.S. Census Bureau, which surveys businesses every two weeks, found that 19.8% of U.S. businesses reported using AI as of May 2026.

That average hides a big split. Nearly 40% of firms in the information sector use AI, and about a third in finance and insurance. Retail sits around 14%. The more telling detail is who’s moving: between December 2025 and May 2026, AI use grew among firms with at least 20 employees, while smaller firms showed no significant change.

Even companies that have adopted AI mostly use it for words. In the Census Bureau’s 2026 survey, 57% of AI-using firms applied it in three or fewer business functions, most often sales and marketing. Writing, coding, analysis and marketing are real uses. But they’re not where most of the economy’s daily operations happen.

Follow the money and you see the gap

Venture funding tells a similar story. One analysis of 2026 deals found that legal, healthcare, construction and insurance captured nearly three-quarters of disclosed dollars going to industry-specific AI companies. Those are big, document-heavy industries where a single customer can be worth a lot, so the logic makes sense.

It also leaves a long list of industries largely untouched: regional transportation, independent property management, the trades, logistics for small and mid-size operators. The operations are complicated, the revenue is real, and very few people are building seriously for them.

This gap isn’t new. Years before generative AI, an industry survey found that 91% of trucking companies with 20 or more trucks used transportation management software, compared with 7% of single-truck owner-operators. The divide between operators with software and everyone running on phones and spreadsheets has been there a long time. The real question now is whether AI closes it or widens it.

They don’t hate technology. They trust what they understand.

The usual explanation is that these businesses are resistant to technology. That hasn’t been my experience. Most operators I’ve met aren’t against software. They’ve been running things a certain way for years, and even when the process is inefficient, they trust it because they understand every part of it.

So when you introduce new software, their first question is almost never about how advanced it is. It’s simpler than that: is this actually going to make my day easier, or am I going to have to change everything just to use it?

That’s a fair question, and a lot of software fails it. Tools built for large companies often assume a dedicated admin, clean data, and weeks of setup. An operator running a busy day off a phone doesn’t have any of that. If adopting the tool feels like a second job, the spreadsheet wins.

The spreadsheet is hiding the real product

The biggest surprise for me has been how much harder it is to automate a manual process than it looks from the outside. You look at a spreadsheet and think, we can turn this into software in a few weeks.

Then you sit with the people who actually use it. That spreadsheet turns out to hold years of small rules, exceptions and workarounds that were never written down anywhere. Things like which driver doesn’t take certain routes, which customer always needs a call ahead, which column means something different on Fridays. None of it is in a manual. It lives in the heads of the people doing the work, and the spreadsheet is just where it leaves a trace.

This is where AI gets interesting, and where I think most people misread the opportunity. The goal isn’t to replace the spreadsheet with a prettier screen. It’s to understand the decisions happening around the spreadsheet, and figure out which of them can be automated and which still need a person.

Traditional software was bad at this. It needed every rule spelled out in advance, so the messy, unwritten parts got left behind. AI is better at working with unstructured inputs like texts, call notes and half-filled forms. That makes it possible, for the first time, to build tools that meet operators where they already are, instead of asking them to change everything first.

What the opportunity actually looks like

I don’t think the winners here will be big platforms that promise to run an entire business. I think they’ll start with one painful workflow and do it extremely well: reshuffling assignments when the day changes, handling the flood of scheduling calls, turning a tenant’s text about a leak into a tracked repair, getting paperwork and invoices out without someone retyping them at night.

Those aren’t exciting demos. But each one gives an operator back real hours, and that’s the only pitch that matters to someone running a busy day.

It won’t be easy, and it’s worth being honest about why. These businesses are careful with money, and they’ve been burned by software that promised a lot and delivered setup headaches. Trust breaks fast: an AI tool that gets a dispatch decision wrong once, at the wrong moment, may not get a second chance. Getting a tool into someone’s daily routine often matters more than which model sits underneath it.

That’s exactly why I think the next wave of useful AI companies will look less like Silicon Valley and more like the businesses they serve. They’ll be built by people who’ve sat next to the dispatcher, watched the spreadsheet in action, and understood why it works before trying to replace it.

The most valuable AI product in some of these industries probably won’t look impressive on stage. It’ll just be the thing that means the dispatcher makes far fewer phone calls, and goes home on time.