Who's Actually Winning the AI Economy (It Isn't Who's Spending the Most)

Most businesses are using AI now, and most of them have little to show for it at the business level. A revenue gap has opened between the companies that rebuilt a real workflow around AI and kept a person in the loop, and the ones still paying for tools that never stuck. The winners are the best implementers, not the biggest spenders, and a lot of them are small, non-tech businesses. The good news for everyone else is that the expensive part of AI is now cheap, and the part that actually wins is a decision, not a budget.

There is a solid case that the boom itself is not a bubble. Exponential View spent months building a bottom-up model of what the world actually pays for AI, and if you follow the market, their full report is a genuinely interesting read. But the size of the boom tells you nothing about whether your business will get anything out of it. That is the question worth answering, so start there.

Who Is Winning With AI, and Who Is Losing

Start with the scoreboard, because it is more lopsided than the tidy adoption numbers suggest. Nearly everyone has AI now. In one survey of 6,000 workers, 87 percent use it and the average person saves around 11 hours a week. And yet only 13 percent of organizations say AI has actually made the company perform better. MIT's team found the same thing from the other side: about 95 percent of enterprise AI pilots return nothing measurable to the bottom line.

It is a bit like a gym in January. Almost everyone signs up, the equipment is packed, and three months later most bodies look exactly the same. The few who changed did not buy more equipment. They changed the routine and kept showing up. AI is having its January right now, at scale, and most of the memberships are aspirational.

The winners are a different story, and you can see them in the numbers. Ramp sees the card and bill payments of tens of thousands of US companies, so it can watch AI spend and revenue at the same time. Since 2023, the top quarter of AI spenders on its platform have more than doubled their revenue. The bottom quarter stayed flat. Sorted by intensity, heavy AI users grew revenue around 27 percent a year while companies with no AI spend grew about 3 percent, roughly the speed of the economy itself. That is more than twenty points a year, and it compounds.

Read that as directional, not gospel. Ramp's base skews toward fast-moving early adopters, so it is not a clean sample of the whole economy. But the shape of it, a small group pulling away while the middle stalls, shows up in every serious dataset on this.

And the winners are not all in San Francisco. The businesses Ramp surfaced were a roofing company in Texas, a window installer in Utah, and a five-person construction firm in Florida that grew 65 percent in a year. These are operators, not labs, running the kind of business that has a truck, a schedule, and a phone that rings.

What Separates the Winners From the Losers

This is where most write-ups stop, right after the “spend more, grow more” chart. That reading is incomplete. The companies that spend the most are usually the ones already doing the hard integration work, and the same research shows that spending without that work returns almost nothing. The budget is a receipt of the work, not the cause of the results.

It is worth being precise about what “AI spenders” even means here. The tool and token bill is the small, visible part. The companies that show up as heavy spenders are usually the ones making the larger investment underneath it, in change management, retraining, and the operational shifts that let them actually run the work differently. The spend tracks revenue because it marks that whole commitment, not because the tokens are magic.

Think about the espresso machine. You can buy the exact one your favorite cafe uses. It will not make you a cafe. The cafe rebuilt its whole morning around that machine: the prep, the sequence, the person who knows when to pull the shot. The machine was the cheap part. AI works the same way. The model is the espresso machine now, and it is nearly free. The winning happens in the workflow you build around it.

The research is unusually consistent on what that looks like. McKinsey found that the single practice most tied to real profit impact is redesigning the workflow around AI, and only about one in five companies actually do it. The rest bolt AI onto the process they already had and wonder why nothing moved. BCG puts numbers on it: roughly 10 percent of the value comes from the model, about 20 percent from the data and tech to deploy it, and around 70 percent from the people and process changes wrapped around it. A separate study of more than 2,000 firms found the ones working from an actual AI strategy, rather than ad-hoc tool adoption, were about twice as likely to see AI-driven revenue growth.

The last piece is control. Klarna automated most of its customer support, watched quality slip, and rehired people into a hybrid model where AI handles volume and humans hold the calls that matter. Human-in-the-loop is a design requirement, not a footnote. It is what keeps the output good enough to trust, which is the only way a workflow survives past the demo. The right people stay in the right seats, and “I want to stay in control” stops being a fear and becomes the operating model.

What the Data Actually Implies

Put the pieces together and the implication is uncomfortable for anyone hoping to buy their way in. Everyone has the tool, so the tool cannot be the edge. The biggest budgets still come up empty most of the time when the work around the tool does not change, so the budget cannot be it either. What is left is the operating layer: knowing which workflow to fix, in what order, with a person on the outputs that carry risk. That is judgment, and judgment does not come from procurement.

It also compounds, which is the part that should worry anyone still waiting for a clearer signal. The gap between a company that redesigned one workflow last quarter and one that did not is small today and structural in a year. The winners are not sprinting; they are putting one workflow into production, then the next, while the tools underneath them get cheaper and better every few months. The lead builds quietly, the way these things do, until it shows up as a pricing problem, or a margin problem, for whoever waited.

Why Small Businesses Are Built to Win This, and Win Cheap

Here is the part almost nobody tells a small-business owner: on this particular field, being small is the advantage.

Start with cost. Large enterprises usually run AI on usage-based API billing, where every token the team spends is metered and the bill climbs with adoption. That is what drags them toward custom builds and a technical team to keep the whole thing standing. A small business does not have to play that game. A flat monthly subscription to the Claude app, between $100 and $200 a month per person depending on how hard you lean on it, includes the usage and needs nothing built or maintained. The most expensive input in the AI economy, frontier-grade intelligence, now costs about the same as a single premium software seat.

Then complexity, which is the real edge. You do not have a legacy system to untangle, a committee to convince, or a procurement cycle to survive. When an owner decides to change how quotes get sent, the decision and the redesign can happen in the same week. A large company needs a steering committee, a pilot, a change-management deck, and a quarter to make the same move, and by then the workflow has drifted again. Size is not the disadvantage on this field. Slowness is, and a small business does not have to be slow.

Add those up and you get the claim that should reframe how a small business thinks about all of this. Thoughtfully applied to a single real workflow, AI can put a small business in the same league as the heaviest AI beneficiaries, at a fraction of their cost. The reason is simple. The expensive things the enterprises are buying, the compute, the custom platforms, the headcount to run them, are not what is producing the revenue gap. The redesign is, and a redesign is available to a fifteen-person company for the price of a subscription and a few good decisions. Fifty-eight percent of US small businesses already use AI, up from 23 percent in 2023, so access is not the differentiator anymore. Sequencing is.

What “Thoughtfully Applying AI” Actually Looks Like

“Redesign the workflow” stays abstract until you watch it collapse a real one. So picture a small sales team.

Today their process looks like most: they run a strong meeting with a prospect in person, shake hands, and drive back to the office. Then the real lag begins. Someone pulls the numbers, builds the quote, routes it for an internal check, and two or three days later the proposal finally lands in the prospect's inbox. By then the room has gone cold. The prospect has taken two competitor calls, half-forgotten the specifics, and the urgency that filled the meeting has quietly drained out through the delay.

Now rebuild that same workflow around AI, with the rep still holding the relationship. The quote gets built live, in the room. The rep talks through what the client needs while an AI workflow drafts the proposal from the conversation, prices it against the current rate card, and formats it into something clean. The rep reads it over, because a human stays on anything that carries money, tweaks a line, and the client leaves with a quote in hand, sometimes a signature, before the momentum ever fades.

Nothing in that requires a data center. It requires one team deciding that the three-day quote gap is the workflow worth killing first, and building AI into it instead of bolting a chatbot onto the side. The payoff is not “time saved” in the abstract. It is a higher close rate, because deals are won in the moment of intent, and faster revenue velocity, because cash that used to arrive in weeks now arrives in days. By the time a competitor's quote lands three days later, the deal is already signed. Do that to the one process that most directly touches revenue and you have moved into the winners' column without spending like them.

It does not have to be sales. The roofing company from earlier did the same thing to a different bottleneck: it stopped letting inbound calls slide to voicemail and rebuilt how they get answered and booked, which is where its money was leaking. Same principle, different workflow. Up close, the winners nearly all look like this. They found the one process quietly losing them money and rebuilt it, with a person still steering.

Where to Start

So the honest advice is smaller than the hype and harder than buying a tool. Start with one workflow, not a platform. Pick the highest-friction, most repetitive thing in your operation, the part that quietly eats a day a week or bleeds a deal a month, and go deep on that one until it runs. Then do the next one. For a lot of operators the blocker is not belief, it is that “I know I should be using AI but I don't know where to start,” and the answer is always a specific workflow, never a shopping list.

For the fractional executives running point across several companies at once, that sequencing judgment is the job. Knowing which workflow to fix first, in what order, with a human on the outputs that carry risk, is worth more than any tool in the stack. Getting in is cheap. Knowing where to point it is the whole thing.

The AI economy question that actually matters is not whether the boom is real. It is whether the next workflow you touch changes how your business runs. Map the workflow first, then build to fit it. That is the line between the companies quietly compounding an advantage and the ones still paying for tools that never stick.

If you have been told AI would change your business and you have not felt it yet, this is the call worth taking. On one free consultation we will map the single highest-friction workflow quietly costing you a day a week, and show you where AI fits and where a person stays on the output. That is the map, whether or not you work with us. Book a free consultation.

Frequently Asked Questions

Ready to find yours? A free consultation maps your highest-friction workflow and shows where AI fits and where a person stays on the output. Book a free consultation.