What Businesses Actually Do With AI — And Why It's Not the Headlines

For two years the headlines have promised the same thing. AI is going to replace everyone. AI is going to transform every industry overnight. AI is an existential threat to your business, your job, and possibly civilization.

Then you walk into an actual company — a logistics operator, a law firm, a manufacturer — and the reality is quieter. Someone uses a chatbot to draft a first version of a client email. The accounting team runs invoices through a system that flags the odd ones. A manager asks an AI tool to summarize a forty-page report into five bullets before a meeting.

That's it. That's the revolution, on most days. And it's worth understanding why the gap between the headlines and the reality is so wide — because the reality is where the opportunity actually is.

What adoption actually looks like

Start with the number nobody puts in a headline. According to the US Census Bureau's Business Trends and Outlook Survey — a biweekly, nationally representative measure of what American firms are actually doing — somewhere between 17% and 20% of US businesses reported using AI in production as of early 2026. Another 20–23% expected to start within six months.

That's not nothing. It's growing fast. But it is a long way from "every business is being transformed." Four out of five firms are still on the sidelines or just stepping in.

Dig into the Census data and the picture gets more specific. Adoption tracks with company size: about 37% of firms with 250 or more employees use AI, versus under 20% of firms with fewer than 20 people. It tracks with sector: the information industry sits near 40%, finance and insurance around 34%, retail trade closer to 14%. And as of November 2025, the survey started measuring AI use across fifteen distinct business functions — finance, HR, customer service, marketing, IT, and the rest — precisely because "using AI" is too vague to mean anything on its own.

The honest summary: adoption is real, it's accelerating, and it's narrow. Companies aren't adopting "AI." They're adopting AI for a specific job in a specific corner of the business.

The 88% that isn't what it sounds like

Here's where it gets interesting. McKinsey's State of AI report from November 2025 found that 88% of organizations report regular AI use in at least one business function — up from 78% the year before. Taken alone, that number sounds like the revolution arrived.

The next number is the one that matters. Only 39% of those organizations report any measurable impact on enterprise profit, and for most of them that impact is under 5%. Only about a third have moved past piloting into actually scaling AI. And the share capturing genuinely significant value — meaningful EBIT impact plus a self-reported "significant" return — is roughly 6%.

So: 88% are using it. About 6% are winning with it.

That gap is the whole story. It's not a technology gap — every one of those companies has access to the same models. It's a focus gap. The 6% aren't using better AI. They're using it more deliberately, on problems narrow enough to actually solve.

The unglamorous truth about what AI does well

When you look at where businesses get real value, the use cases are almost boring. That's the point.

The reliable wins share a structure: a high-volume, well-defined task with structured inputs and a predictable output. Drafting a first version of a routine document. Summarizing long material into a short brief. Answering the same fifteen customer questions that make up most of the support queue — around 41% of companies now use AI to help generate customer responses. Pulling data out of one system and reconciling it against another. Flagging the invoice, the order, or the record that doesn't look right so a human can look at it.

None of that is "AI replaces your team." All of it is "AI removes the part of the work that was never worth a human's attention in the first place." The first draft still gets edited. The flagged exception still gets a human decision. The summary still gets read by someone who's accountable for the meeting.

This is the part the headlines miss entirely, because "software quietly drafts a first pass and a person finishes it" doesn't generate clicks. But it's where the money is, and it's why the fear is mostly misplaced. The businesses getting value from AI aren't firing their people. They're handing the mechanical work to a machine and pointing their people at the work that actually needs judgment. We've written before about why AI replaces tasks, not jobs — the data on how companies actually deploy it backs that up.

Why narrow beats broad, every time

The instinct, especially from the top of an organization, is to go big. Announce an AI initiative. Stand up a committee. Try to transform everything at once.

It's the most reliable way to end up in the 88% that shows nothing for it.

Broad transformation fails for unsexy reasons. The data isn't clean enough to feed every process at once. The workflows aren't documented well enough to automate in bulk. The edge cases aren't understood until something is live and handling real volume. A company that tries to automate ten things simultaneously discovers all ten sets of problems at the same time, gets overwhelmed, and quietly shelves the whole thing.

The companies in the winning 6% do the opposite. They find one process — invoicing, intake, reporting, reconciliation — that runs constantly and eats real hours. They automate that one thing properly, with the inputs constrained and the outputs validated and a human reviewing the exceptions. They measure it against their own baseline. And then, with a working system and a real number in hand, they pick the next one.

That progression is the entire game. It's also why we built our Five Levels of AI Maturity framework around movement, not leaps — most businesses sit at Level 1 or 2, where someone has experimented with a tool but there's no structure around it. Getting to real, autonomous output isn't a single jump. It's a sequence of narrow wins.

AI isn't a strategy. It's a tool.

Most of the fear, and most of the wasted money, comes from a single category error: treating AI as a monolithic thing you either adopt or get crushed by.

You don't adopt electricity. You wire it into specific things — the lights, the machines, the heat — each one a discrete decision with a discrete payoff. AI is the same. It's not a destination you arrive at. It's a capability you point at problems, one at a time, keeping the ones that work and dropping the ones that don't.

Framed that way, the scary version evaporates. You're not betting the company on a transformation. You're asking a much smaller, much more answerable question: which single process in my business is repetitive enough, high-volume enough, and well-defined enough that a machine should be doing it instead of a person?

That question has a real answer for almost every business. And answering it well — picking the right first process, building it so it's reliable, measuring what it returns — is worth far more than any amount of strategy-deck ambition about transforming the enterprise.

The headlines will keep promising the revolution. The companies actually getting value will keep doing the quiet, specific, unglamorous work of automating one process at a time. The second group is the one worth copying.

Related reading

• The Five Levels of AI Maturity — the progression from experimenting with a tool to running real automated output.
• Top Five AI Myths, Debunked — the fears that keep businesses on the sidelines, and what the evidence actually says.

---

Not sure which process in your business is the right place to start? Take the free AI Maturity Assessment — 25 questions, your score, and a roadmap — or book a discovery call.