The Maze: Ecommerce can transform who wins without making the whole retail industry richer. In the United States, online sales rose from 0.9% to 16.4% of retail between 2000 and 2025 in a BCG comparison shared by Gregory Emerson. Goods consumption fell from 23.9% to 21.0% of gross domestic product (GDP). The tension matters for artificial intelligence: a useful technology can become essential to compete even when its benefits do not become extra industry profit.
A bigger digital channel need not mean a bigger spending share. Ecommerce gained 15.5 percentage points of retail sales across the two endpoints. The Census release confirms the 16.4% annual share for 2025; that is different from its quarterly figures. Goods consumption moved the other way relative to GDP, down 2.9 points. These measures have different denominators. They show a shift in the route to customers alongside a smaller goods share of the economy. They do not show that households spent fewer dollars, or that ecommerce reduced consumption. A larger economy can support more spending while goods account for less of it.
Customer value can grow without appearing in retailer margins. Emerson frames the ecommerce payoff as consumer surplus: benefits beyond the price customers pay. Easier comparison, wider choice and time saved are plausible mechanisms. They can improve the shopping experience while competition forces retailers to pass savings on. The BCG margin comparison shows only a modest movement between early adoption in 2001–05 and the post-pandemic period in 2022–25. It provides neither precise endpoint labels nor the retailer sample. That is an illustration of the argument, not proof that every cost saving reached shoppers. Stable margins cannot tell us what profits would have been without the technology.
Aggregate results hide radically different company outcomes. The BCG comparison places Best Buy's $12.5 billion FY2000 revenue beside $41.5 billion in FY2025, while Circuit City ends in liquidation in 2009. These are historical examples of divergence, not a controlled test of digital investment. Best Buy's results independently confirm $41.528 billion in FY2025 revenue and a 39.5% online share of domestic revenue in its fourth quarter. That mix belongs to one quarter and one segment, not the entire company year. Revenue is also not market share. The useful distinction is between an industry's average performance and an individual retailer's ability to keep customers.
An AI business case needs a competitive counterfactual. A return on investment calculation normally compares benefits with spending. But keeping revenue that would otherwise disappear also has value. If rivals use AI to answer customers faster or lower the cost of serving an order, standing still may weaken a retailer even when everyone ends up charging less. That is an inference from competitive economics, not an AI result measured by these retail endpoints. It still demands tests: compare customer retention, service cost and contribution profit with a credible alternative. Otherwise, a survival argument can become a blank cheque for projects that merely look busy.
Why it matters: Retailers should separate three questions: did customers get more value, did the firm gain or defend sales, and did the industry earn more? Those outcomes can diverge. For AI, count defensible avoided losses alongside new revenue and savings, then test them against an actual baseline. Watch whether a better customer experience holds profitable repeat business as competitors copy it. Adoption alone proves that a tool spread. It does not prove who kept the economic benefit.
Images: Cover AI-generated
