Most AI case studies you have read come from banks, software companies and consultancies. That is not a coincidence. That is where AI is actually being used.

In May the US Census Bureau put numbers on it. Its Business Trends and Outlook Survey asks a sample of firms every two weeks whether they used AI in any business function. In early May, 19.8% said yes. Information was at 39.7%, finance and insurance at 33.9%, retail around 14%.

A few days later the Minneapolis Fed sliced it further. Fewer than 10% of firms in agriculture, transportation, accommodation and food services, and construction reported using AI. Health care was around 20%, manufacturing around 12%.

The bar here is low. Any AI use in any business function in the last two weeks counts. Whole sectors still do not clear it.

Why these sectors trail

The Fed gives three reasons: how digitized the sector is, how much data is available, and how complex the tasks are.

All three say the same thing. Today’s AI works on text, numbers and images already sitting in a computer. A bank analyst spends the day in front of those. A site foreman spends the day in front of a concrete slab. A driver, in front of a highway. A housekeeper, in front of a room. None of those jobs is going to change because of a language model this year.

Stop there and the conclusion is that AI is not for these industries yet. I think that is wrong.

The office behind the operation

Take a mid-sized contractor. The job site is physical. Before it, though, there are bids, estimates with hundreds of line items, drawings going back and forth, supplier quotes. During the job there are daily logs, change orders, payroll, client reports. After it, invoicing, retainage and collections.

That whole half is paper and screens. In a lot of companies in Latin America, where I work, it lives in email, spreadsheets and WhatsApp threads.

A hotel works the same way. The room is physical. Rates, bookings, the same guest question in three languages, reconciliation with online travel agencies, the shift schedule. All of that is already digital. In trucking it is quotes, route assignment, dispatch, bills of lading and collections.

So I read the adoption gap as a head start. In banking, your competitors already automated the back office and catching up is expensive. In hospitality or transport, most have not started, and the office behind the operation holds exactly the kind of work where AI has already proven itself.

One caveat I owe you. All of this is US data. I do not know of a comparable public measurement for Costa Rica or Central America, so the regional part of this post is my opinion.

The prerequisite that is usually missing

There is a condition, and it is the one the Fed data hints at. AI needs the information to be recorded somewhere it can read. If the quote lives in the operations manager’s head and the job status is known by calling the foreman, there is nothing to point AI at.

In that case the first project is a recording project. Quotes in one format, bookings in one system, change orders in one place. Less exciting, and it is what makes everything after it possible.

Yesterday I wrote about Harvard and INSEAD’s experiment with 515 startups, where the group taught where to put AI earned 1.9x the revenue with the same tools. In low-adoption sectors that map matters even more, because there are no competitor examples to copy.

How IQ Source approaches it

When a company from one of these sectors comes to us, we split the physical operation from the office that runs it and check how much of that office is actually recorded. If it is, the work goes through AI Maestro, mapping the flow, finding where it jams, deciding which AI use cases are worth building. If it is not, it starts with digital transformation, putting systems in place so the data lands where AI can read it later.

Here is an exercise for this week. Count how many quotes, bookings or dispatches went through WhatsApp or a phone call without ending up in any system. That number tells you whether your first project is AI or recording. Send it to me and I will tell you which one I would start with.

Review my back office with IQ Source