Better matching produces more transactions. It does not guarantee better economics.
Here is a small puzzle that tells you a lot about where AI is taking business. Walmart works with tens of thousands of suppliers. For most of them, no Walmart buyer ever sat down to negotiate individual terms. It wasn’t that anyone thought these suppliers didn’t matter. The arithmetic simply didn’t work: the gain from negotiating with each one was smaller than the cost of a human doing it. So they all got the standard deal.
Then Walmart handed those negotiations to an AI. I will come back to what happened, because it is a wonderful illustration of the promise of what people now call the agentic economy. Agents can search more options, surface demand nobody saw, negotiate, assemble offers on the fly, and make transactions worthwhile that human time and complexity used to rule out. All of that is plausible, and some of it is already happening.
But here is what worries me. A company can become dramatically better at finding and serving demand and still destroy value. That is where most of today’s conversation about agentic commerce stops too early. It treats more transactions as the goal. The goal is economic profit.
Why agents change the arithmetic
Ronald Coase explained back in 1937 why firms exist at all: using the market is costly. Finding counterparties, negotiating, and enforcing agreements all take time and money. [1] Sangeet Paul Choudary has applied that logic to AI with real clarity. Agents cut the cost of discovery, negotiation, and coordination, and when those costs fall, deals that were never worth organizing suddenly are. In his words, “commercial arrangements that were previously too expensive to negotiate became economically worthwhile.” He reports that Honeywell has already completed more than 2,500 autonomous supplier negotiations covering over $500 million of spend. [2]
Which brings me back to Walmart. Its pilot with Pactum’s negotiation agent on that long tail of suppliers, documented in Harvard Business Review in 2022, closed agreements with roughly two-thirds of the suppliers it approached, with modest average savings and longer payment terms. [3] The savings rate is not the point. The point is that a negotiation that made no economic sense with a human on each side became worth doing once the cost of negotiating fell close to zero.
Execution is becoming abundant in the same way. BCG describes a consumer goods company that used to produce around 500 marketing assets a year and now needs something on the order of five to six million. The unit cost has collapsed along the way: a personalized product shot for a beauty item that cost $4 to produce two years ago now costs four cents. [4] Five hundred to five million, at four cents a shot! When execution stops being scarce, it stops being a source of advantage. What stays scarce is the judgment about where to point it.
All of this answers whether a transaction can happen. It leaves the more important question wide open: should it?
That sounds strange in a business culture built around conversion. It shouldn’t. Not every customer should be won, not every demand space should be served, and not every transaction creates value.
Winning demand can destroy value
Let me make it concrete. A retailer serves a particular demand space very well and earns $40 million a year in after-tax operating profit from it. Everyone is pleased. But serving that demand ties up $500 million of capital in inventory, fulfillment capacity, technology, and working capital. At a 9 percent cost of capital, the capital charge is $45 million.
$ millions
After-tax operating profit 40
Capital employed 500
Capital charge at 9% 45
Economic profit −5
The company found the demand, served it, grew revenue, and reported a healthy operating profit. It also earned 8 percent on capital that costs 9 percent, and destroyed $5 million of value in the process. Now give that company an agent that is brilliant at finding and closing this kind of demand. It will simply destroy value faster.
I wish this were an edge case. McKinsey’s analysis of economic profit for nearly 3,000 large nonfinancial companies found that the top fifth generated $677 billion of economic profit, while the bottom fifth destroyed nearly $411 billion. The 60 percent in between generated a little over $29 billion combined, about $17 million each. [5] In other words, most large companies barely earn their cost of capital. McKinsey Global Institute’s later work on nearly 6,000 firms found the same steep curve: the top 10 percent capture about 80 percent of economic profit. [6]
Finding demand faster does nothing to that arithmetic. Choosing better demand can.
A new consumer study adds a twist that matters for where the money goes. AlixPartners surveyed more than 9,000 American shoppers for its 2026 Fashion Consumer Sentiment Index, and it expects agentic commerce to cleave fashion rather than hit it evenly. At the top, luxury stays brand-led. At the bottom, the replenishment buys, the black leggings under $30, will happily be handed to an agent that decides on specification and price. Most brands live in the middle and will have to manage both. [7] Julie Bornstein, who founded the fashion shopping start-up Daydream, made the same point to The Economist: chatbots work well for products with comparable specifications, like vacuum cleaners, but in fashion “it’s not about the specifications of the product. It’s about the feel of the product, the vibe of the product.” [8] Seen through demand spaces, this is a capital allocation question. In the demand spaces an agent treats as a commodity, money spent building preference may never earn its cost of capital. In the ones where the brand still decides, underspending is the bigger risk.
Not everyone buys this. David Rissmann, who leads strategy at the beauty retailer flaconi, points out that in the same survey only 10 percent of shoppers start their discovery with AI, and that among Gen X and Boomers easy human support beats shopping technology 51 percent to 23 percent. His sharper objection is about friction. A reorder is already the purchase with the least effort left in it; at flaconi, rebuying your usual fragrance takes a few taps in the app. An agent earns its place by removing effort, and here there is little left to remove. [9] I think he is right about the reorder, and it is Coase’s point from earlier in a different guise. Agents create the most value where transaction costs are high, which is why the long tail of supplier negotiations moved first, and the least where costs are already close to zero. It also tells you how to treat the split AlixPartners predicts. Nobody knows when it arrives, so it is not something to bet the company on. It is something to prepare for cheaply, with the option kept open, a distinction I come back to below.
What should the company’s agent optimize?
This becomes urgent the moment companies deploy their own agents. Tell an agent to maximize conversion, and it will get very good at closing deals. Tell it to maximize revenue, and it will find volume. Tell it to maximize customer satisfaction, and it may happily over-serve customers at a loss. None of these maximizes economic value, and an agent will chase whatever objective you give it with a consistency no sales force ever had.
The agent needs an economic mandate, and that mandate has to come from strategy. Which demand spaces are attractive? What will people pay in each? What margin do we need, how much capital is tied up, and which capabilities do we have to build? What does a win here cannibalize elsewhere in the portfolio? What risk are we taking on, which future are we betting on, and where should the agent simply say no?
This is where Demand Architecture, the Brand System, and economics have to come together. Without that, companies are in for an uncomfortable discovery. Their AI will become extraordinarily good at executing strategies that should never have been chosen.
BCG gets halfway there. In a recent piece for CEOs, its authors argue that the question for marketing budgets “is no longer simply whether to spend more or less, but where the next dollar will generate the greatest return.” [10] I agree, but I would push the question further upstream. Where the next marketing dollar works hardest is still a question asked inside the demand a company already serves. The prior question is which demand deserves capital at all, and not just marketing capital, but the inventory, the fulfillment, the technology, and the people behind it. Get that wrong, and the most efficient marketing in the world simply helps you destroy value more efficiently.
Today’s demand is not tomorrow’s
There is a second reason not to optimize against the demand the agent sees today. The structure of demand changes, and it changes faster than category data show. People are drinking less alcohol. Working patterns are shifting. Health habits are changing. New technologies create substitutes, regulation changes the economics, and new categories swallow old needs. A demand space that looks attractive today can look very different in five years.
Fashion is getting a live lesson in this right now, and the most striking finding in the AlixPartners study has nothing to do with AI. With about one in nine Americans now using GLP-1 medications, the share of activewear shoppers who call consistent fit and sizing an important reason to buy nearly doubled in a year, from 29 to 54 percent, while the importance of “inclusive product options” dropped by a third. Some retailers are already cutting plus sizes; Reformation’s plus-size assortment is down 60 percent. As the authors put it, sizing data is lagging “a population whose bodies are changing faster than any product cycle can track.” [11] (And there is the economic sting: brands that keep planning to a three-year-old snapshot have working capital sitting in inventory in the wrong sizes.) That is the kind of tide I mean. The category looks the same, and the demand underneath it has moved.
Most growth strategies still assume a fundamentally stable market in which growth is a matter of winning share. But share gains explain only a small slice of corporate growth, about 7 percent in McKinsey’s analysis. [12] The bigger opportunities, and the bigger threats, come from shifts in the structure of demand itself. That is why I developed what I call the Structural Change Index. The question is not only how fast a market is growing, but how much the structure underneath it is moving. Are new demand spaces appearing and old ones shrinking? Are categories converging? Is willingness to pay migrating? Are substitutes coming in from outside the category? In our work, we keep finding that even categories that look perfectly calm on the surface have strong tides running underneath.
A company can gain share in a market whose economics are quietly deteriorating. It can also lose share in an old category while building a much stronger position in an emerging demand space. Share alone can’t tell you which of the two you are in.
Working back from the future
Consider a spirits company with strong economics today in a demand space built around social celebration. Several futures are plausible. People may drink less but trade up. Social occasions may come roaring back. Moderation may become the norm, with functional beverages or cannabis taking over part of the same occasion. Each future implies different volumes, different prices, different price tiers worth competing in, and a different balance between bars and retail.