AI For Businesses That Already Work

Key Takeaways

• Adoption is not the problem any more. In several industries most companies already use AI in some form.
• The results are not matching the expectation, and the gap is measured, not anecdotal.
• Almost nobody measures return. Which is why the disappointment is vague rather than specific.
• The value lands in the boring middle, in follow up, admin and reporting, not in the impressive demo.
• In Latin America the constraint is not enthusiasm. Budgets went up. Integration did not follow.

Most AI writing is about companies that sell software. I spend a good part of my week with companies that sell cars, meals, legal advice, dental work and industrial parts. Businesses that already work, that made money before any of this, and that have a real cost of being wrong.

The data on what is happening inside them is finally good enough to argue with.

Adoption is already here, and lopsided

The US Census Bureau's business survey put AI use at 17 to 20 percent of firms through the first half of 2026, but the interesting part is the split by size: 37 percent among firms with 250 or more employees, under 20 percent among firms with 1 to 4. Use rose for firms above 20 employees and did not change significantly below that. (US Census Bureau, May 2026)

Then look at sector surveys and the numbers jump. Eighty two percent of auto dealers report using AI. (Cox Automotive, August 2026) Eighty one percent of physicians report awareness or use, up from 38 percent in 2023, averaging 2.3 use cases each. (AMA survey, March 2026) Forty eight percent of home service trades businesses actively use it. (Housecall Pro, June 2026) In professional services, organisation wide use nearly doubled to 40 percent in a year. (Thomson Reuters, February 2026)

Restaurants are the outlier, at 26 percent, with most of that in marketing rather than operations. (National Restaurant Association via Restaurant Dive, February 2026)

So the story is not that traditional businesses refuse to adopt. Most of them already did something.

The expectation gap is the real finding

The single most useful number I have read this year is from the dealership survey. Sixty nine percent of dealers expected AI to grow sales. Among those using it, only 22 percent got the sales growth they expected. (Cox Automotive, August 2026)

That is not a technology failure. That is a promise failure.

The professional services data says the same thing from the client side: 91 percent of professionals say their organisation falls short of what AI could deliver, only 6 percent of providers meet client AI expectations, 78 percent of clients now consider AI enabled quality essential, and 32 percent would reconsider the relationship within a year. (Thomson Reuters, June 2026)

And underneath both, the number that explains the fog: only 18 percent of professional services organisations track return on their AI tools, unchanged year over year. (Thomson Reuters, February 2026)

If you do not measure it, you cannot be disappointed precisely. You can only be disappointed generally, which is worse, because it does not tell you what to fix.

Where the value actually lands

Look at what the same surveys say people use it for, and the pattern is consistent across industries that have nothing else in common.

Dealers: automating routine tasks 40 percent, customer follow up 40 percent, content 38 percent. Trades: customer communication 52 percent, estimates and quoting 51 percent, planning 43 percent, with more than one in four AI users saving six or more hours of admin a week. Real estate agents: listing descriptions, social posts and email, with 68 percent saving at least an hour a week. Small businesses generally: marketing, customer service, bookkeeping. (Intuit QuickBooks, May 2026)

None of that is the demo anybody was shown. All of it is the middle of the business: the follow up that does not happen, the quote that takes two days, the report somebody assembles by hand on Friday.

That is the honest pitch for this kind of work, and it is the one we lead with at Unbound Growth Partners. Not a transformation. A specific queue of repetitive work, removed, with the hours counted before and after.

Latin America, specifically

The regional picture is not a lack of interest.

In Colombia, 61 percent of surveyed leaders call AI a strategic priority and 71.9 percent raised their AI budget in the last year, while only 37.7 percent have integrated it meaningfully into daily operations. The maturity distribution is brutal in its honesty: 58 percent exploring, 35 percent building, 0.2 percent leading. The top barriers are cybersecurity, infrastructure, talent and data quality, in that order. (Endeavor Colombia AI Pulse via Colombia One, August 2026)

Regionally, a World Economic Forum analysis found only 23 percent of Latin American organisations generating any economic value from AI, 6 percent capturing significant value, and 59 percent of small and mid sized firms reporting no impact at all. (WEF with McKinsey, February 2026)

Budgets went up. Integration did not. Anyone who has done implementation work here knows why, and it is rarely the model: the data lives in WhatsApp, the process lives in one person's head, and the system of record is a spreadsheet that person maintains.

How we actually run these engagements

1. Pick one queue. Inbound leads with no follow up. Quotes that take two days. Month end reporting. One, not five.
2. Count it before. Volume, hours, conversion, response time. If nobody can produce those numbers, producing them is the first deliverable, and it is usually worth the fee on its own.
3. Automate the middle, not the edges. The first contact and the final decision stay human in most of these businesses. The retrieval, drafting, routing and logging in between is where the hours are.
4. Put a person on the exceptions from day one. Every one of these businesses has a case where being wrong costs a customer or a licence. Route those to a human on purpose, not by accident.
5. Report the same numbers every month. The ones from step two. This is the discipline that separates a project that renews from a project that quietly stops.

What I tell owners who are sceptical

You are not behind. Most of your competitors bought something, used it for marketing copy, and never measured it. The gap between having AI and getting value from it is currently enormous, and it is not closed by better models.

It is closed by picking one expensive, repetitive, well understood process, and doing the unglamorous work of making it cheaper, with the numbers to prove it.

That is a much smaller promise than the one the market is making. It is also the one that survives contact with a real business.