By 1900, electric motors were providing less than 5% of the mechanical drive power in American factories. Tim Harford, retelling the economic historian Paul David's explanation for the slow payoff, wrote that electricity "triumphed only when factories themselves were reconfigured."
Most companies are in their own 1900 with AI. They have bought the motor. An AI strategy is a decision about how the business should work now that these tools exist, followed by the harder job of rebuilding the work around that decision. A stack of subscriptions is what a company has before it makes that decision, and plenty of companies are stopping there.
In a McKinsey survey of 1,719 professionals and business leaders worldwide, reported by Fortune this month, nearly nine in 10 said their organizations regularly use AI in at least one business function. Just 37% said it was having a meaningful impact on earnings before interest and taxes, unchanged from a year earlier. Yet 80% said AI had improved their individual productivity. On the respondents' own account, the gains are landing on individual desks and stalling before they reach the income statement.
The companies getting paid for it look different. Among the small group McKinsey counts as high performers, nearly three-quarters have redesigned their workflows, up from 55% a year earlier. They changed the factory floor.
Small companies aren't catching up. The Census Bureau found that business AI use hovered between 17% and 20% from December 2025 to May 2026, and that use rose at firms with at least 20 employees while firms with fewer than 20 didn't change significantly. A Census working paper found that firms with AI built into more of their operations tend to perform better. That doesn't prove the AI did it, and better-run companies may simply build it in more widely. I'd still take the hint.
The clearest test so far comes from Denmark. Economists Anders Humlum and Emilie Vestergaard studied Danish workers and found no effect on earnings or recorded hours two years after ChatGPT's launch, with estimates precise enough to rule out effects larger than 2%.
The best argument against me is that tools alone do work. In a study of 5,179 customer support agents, an AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and low-skilled workers. In an experiment with 453 professionals, ChatGPT cut the time it took to complete writing tasks by 40%. MIT's NANDA research even found that buying AI tools from specialized vendors or partnering with them succeeded about 67% of the time, while internal builds succeeded only one-third as often. For a small company without its own engineers, that reads like permission to keep buying.
Those results are real, and an owner should want them. They're also measured one task at a time. A support desk that resolves 14% more issues an hour on average has a better hour. Whether that turns into profit depends on what the company does with the time. It might take more calls with the same staff, or it might move its fastest people to the accounts that need them most. Across the economy, the St. Louis Fed's tracking shows the share of work hours saved by AI grew from 1.6% to 2.2% between roughly the third quarter of 2024 and the second quarter of 2026. In a 40-hour week, 2.2% is less than an hour, and an hour nobody assigns to anything never reaches the income statement.
NANDA's own explanation points the same direction. Generic tools, the report's lead author told Fortune, stall "since they don't learn from or adapt to workflows." Buying is fine. The redesign still has to happen, and a subscription doesn't come with one.
A small company has one advantage here. The people who would have to agree to a new workflow can fit around one table, and the owner is usually one of them.
Pick the process that eats the most of your own week. Quoting a job, scheduling crews, chasing invoices, and answering the same customer question for the fortieth time are all good candidates. Write down every step and who touches it. Then mark the steps that exist only because a person had to carry information from one place to another, or had to remember something nobody wrote down. Those are the steps worth handing to software first. The rest of the design is deciding what the freed-up people do next, and that part is a management decision.
Redesigning the work also makes a business easier to sell, a case I made in an earlier piece on the SBA's new acquisition rules. A process that lives on paper and in a system can outlast the person who built it.
RVRBN builds AI systems for small companies, so we have an interest in owners thinking about it this way. Nothing above requires hiring us. Do the redesign before the next subscription.
The motor was for sale in 1900. The payoff went to the owners who redrew the floor plan.
If you want to talk through what this means for your company, get in touch.