Picture three regional distributors that compete for the same customers. In the same quarter, all three license the same AI assistant from the same vendor and point it at the same work: quotes, order entry, customer email, month-end close. Each one takes roughly 15 percent out of the cost of that work, and each one tells its board about it. A year later, a large customer asks all three to sharpen their pricing, and because all three can now afford to, they do. The savings are real. They now belong to the customer.
The savings are real. They now belong to the customer.
Now picture a fourth distributor in the same market. It bought the same tools, took the same savings, and then used the time and the data to build something its customers could not get elsewhere: a quote that arrives while the buyer is still on the phone, a reorder service that knows a customer’s usage better than the customer does, a view of stock across suppliers that saves a buyer a day of calls. Its prices fell too. Its margin did not, because its customers were choosing it for something other than price.
This article is about the difference between those two outcomes, and why we think it is the most important strategic question in AI for a mid-market company.
The Tool Is Not Scarce
Efficiency built on a widely available tool is, almost by definition, available to your competitors. Microsoft reported more than 30 million paid Microsoft 365 Copilot seats in July 2026. The Census Bureau’s business survey found that 19.8 percent of US businesses were using AI in May 2026, rising to 37 percent among firms with 250 or more employees. A Federal Reserve analysis this spring estimated that about 41 percent of workers used generative AI for their jobs by late 2025, and that about 78 percent of the labor force works at firms that have adopted AI. Each measure uses a different definition, and none of them is a measure of skill. But the pattern is plain: the tools are everywhere, and whatever they make easier is becoming easier for everyone at once.
An Old Argument, Running Faster
None of this is a new idea. Michael Porter made the case thirty years ago this month, in his Harvard Business Review article “What Is Strategy?” in the November and December 1996 issue. Operational effectiveness, doing the same things better than rivals, is necessary, he argued, but it is not strategy, because best practices spread quickly and competitors copy them. Companies that compete only on efficiency end up, in his phrase, in “a series of races down identical paths that no one can win.” In 2003, Nicholas Carr pushed the same logic onto technology itself in “IT Doesn’t Matter,” arguing that as information technology became ubiquitous and cheap, it stopped being a source of advantage, because advantage comes from what is scarce.
Companies that compete only on efficiency end up, in his phrase, in “a series of races down identical paths that no one can win.”
Our view is that both were right about efficiency and that AI makes them more right. A best practice used to take years to spread through an industry. An AI capability that a vendor ships to all of its customers on the same Tuesday spreads in weeks.
Where the Savings Go
When everyone in a market gets cheaper at the same time, competition hands most of the savings to customers. The economist William Nordhaus estimated in a 2004 working paper that innovators kept only about 2.2 percent of the total value their innovations created in the postwar US economy, with most of the rest captured by consumers and other stakeholders. The same pattern is showing up in AI. A Stanford Digital Economy Lab working paper published in April 2026 estimated that generative AI was delivering about $172 billion in US consumer surplus by early 2026, well above the revenue the AI companies themselves were earning, and concluded that consumers were capturing most of the gains. That estimate rests on what users say the tools are worth to them, and the direction matches Nordhaus.
We have described this pattern in our own framework, the Shapes of the AI Dividend: in a Rising Bar market, AI becomes the price of staying on the field, and the dividend flows to customers unless a firm does something to keep it. Our earlier piece on maximizing your AI investment made the same warning in shorter form: productivity gains are becoming table stakes.
Efficiency Still Matters
This is not an argument against efficiency. Porter called it necessary, and a company that falls behind its rivals on cost loses whether or not it has a strategy. The question is what you do with the savings.
The most useful evidence we know on that question comes from recessions, when every company is under pressure to cut. Ranjay Gulati, Nitin Nohria, and Franz Wohlgezogen studied 4,700 public companies across three recessions and reported in the Harvard Business Review in 2010 that only about 9 percent flourished afterward, beating rivals by at least 10 percent in both sales and profit growth. The companies that focused mainly on cutting costs had the lowest chance of pulling ahead, about 21 percent. The companies that cut selectively, mostly through efficiency, while investing in research, marketing, and assets had the highest, about 37 percent. Recessions are a special case, and the study shows association rather than cause. But the lesson carries: savings that fund something new outperform savings that are simply banked or competed away.
What Compounds Instead
Revenue built on technology tends to last longer than efficiency built on technology because it depends on things a competitor cannot buy from the same vendor: your customers, your data, and the way the two work together. James Bessen’s research, published in the Journal of Law and Economics in 2020 and in his book The New Goliaths, links proprietary systems, the custom software firms build for themselves rather than buy, to rising industry concentration and to the growing advantage of industry leaders. Proprietary is the key word. The advantage came from systems competitors could not license.
There is also evidence that markets reward growth more than margin, at least for some companies. McKinsey’s 2014 study of about 3,000 software and online service companies found that for firms under $4 billion in revenue, improving revenue growth drove about twice as much gain in market value as improving margins. That was one industry, studied by a consultancy, and it may not carry to a distributor or a services firm. We cite it as consistent with the argument, not as proof of it.
Our own judgment is that revenue capabilities compound for a simple reason: they get better with use. A pricing model learns from every quote. A reorder service learns every customer’s rhythm. A retention program learns who responds. An efficiency gain, once captured, mostly stays the size it was.
An efficiency gain, once captured, mostly stays the size it was.
Proof From Our Own Work
Two of our published cases show both kinds of gain side by side. In the easy button case, a commercial real estate capital markets firm reported $4.5 million a year in efficiency savings from an AI model that matched properties to likely buyers, and $3.2 million in revenue from a six-point gain in client retention. In our regional brokerage case, a firm reported $3 million in efficiency gains and $5 million in attributed new revenue over 24 months on a $1.5 million investment, with no AI model at all; the revenue came from data the firm could put in front of a prospect.
Both firms banked the efficiency, and they should have. But in our reading, the revenue is the part a competitor could not have bought the next quarter, because it rested on the firm’s own client history and its own way of using it. These are the firms’ own figures, they are two cases, and we have not audited them. We offer them as illustrations of the pattern, not as a measurement of it.
How to Tell Which Kind of Gain You Are Buying
Before funding the next AI project, we suggest asking three questions about it, and these are our judgment rather than a standard. First, could a competitor buy the same result from the same vendor next quarter? If yes, it is an efficiency gain worth having, but not an advantage. Second, does the gain show up in a customer’s decision to choose you or stay with you? If not, it is likely to be competed away. Third, does it get better the more you use it, because it learns from data only you have? Gains that pass all three tests are the ones that compound.
This is also how we weigh projects in Project Vista: a project that builds a capability later projects can use earns credit for more than its own return. For specific revenue-side uses and how to test them in a quarter, our recent piece AI That Sells lays out five.
What This Will Not Do
The opening scenario is a picture, not a case. Porter and Carr made arguments, not measurements. The recession study covers recessions, the growth study covers software companies, the consumer-surplus estimate rests on stated preferences in a working paper, and our two cases are self-reported. Our reading of Bessen comes from his published paper and summaries of his book, not from the full book. None of this proves that any particular revenue project will pay. What the evidence supports is a direction: treat efficiency as the floor and the funding, and look for the gains a competitor cannot buy.
Where to Start
List your current AI projects in two columns: the ones that make existing work cheaper and the ones that change what a customer gets. If the second column is empty, that is the gap. If you would like an outside view, our independent assessment can sort your portfolio the same way. If you would rather talk it through first, you can reach us through the contact form on our About page.
