Allbirds put its sneakers on Amazon in 2023, and CEO Joey Zwillinger frames the move not as a betrayal of direct selling but as the plan he wanted all along: “We never believed that direct would work to meet the ambition of the scale we thought we could achieve.” That a brand which helped define digitally-native retail now says its founding channel couldn’t scale on its own is a data point in itself. Three years later, the concession looks less like one company’s story and more like an early signal of a shift across the category: the cost of reaching a customer anywhere else has climbed past what direct selling can absorb, and Amazon’s AI-driven discovery has become the cheapest route left to a buyer.

The direct-to-consumer thesis of the 2010s treated the customer relationship itself as the asset: the first-party data, the fatter margin, and the loyalty a marketplace would otherwise capture. Selling on Amazon meant handing over purchase history and pricing a product beside Amazon’s own private label. As late as 2020, the holdouts were a named group: Glossier, Warby Parker, Allbirds, and Away kept their full catalogs off Amazon, while Casper had already begun selling there and Vuori, though it stayed off the marketplace itself, ran ads timed to Prime Day to catch shoppers’ attention elsewhere. Owning the customer was supposed to be worth more than the reach a brand gave up by staying off the marketplace. For most of the 2010s, the bet was affordable, because acquisition elsewhere was cheap.

The bet worked only as long as that assumption held, and it broke as acquisition on Meta and Google stopped being underpriced. The cost of acquiring an e-commerce customer has risen 222% over the past decade, a rise steep enough to erase the margin the direct model was built to protect. Brands now lose an average of $29 on every new customer before that buyer returns to shop again. Owning the relationship still has value; the problem is that filling the channel got expensive, and a channel a brand can’t afford to fill stops functioning as an asset.

While the cost of the owned channel climbed, Amazon rebuilt the one it controls. Its search stopped behaving like a keyword index and started inferring intent: COSMO, the commonsense knowledge graph Amazon deployed across search relevance and recommendation, maps a product to the occasion and need behind a query rather than the words typed into it, and the company reports gains from live A/B tests. Above it runs Rufus, the shopping assistant more than 250 million people used over the past year, since renamed Alexa for Shopping. Shoppers who engage it are over 60% more likely to finish the purchase in that session, because the assistant reaches them after the intent to buy has formed. That is the mechanism the open web cannot match: Amazon answers a question from someone already reaching for a card, instead of buying attention and hoping it converts.

A shopper who asks Amazon’s assistant for a waterproof running shoe is closer to buying than a stranger scrolling past an ad for one, and far cheaper to reach.

The pattern shows up sharpest in a weak market, and apparel is currently that market. U.S. spending on clothing declined modestly in 2025, through November 30, and underperformed consumer spending overall, the kind of softness that turns acquisition cost from a line item into a survival question. Allbirds’ third-quarter revenue fell 23.3% year over year, a decline the company’s own filing attributes to store closures and the unwinding of international distributor deals, not to acquisition spend. The filing’s explanation is not the whole story.

Allbirds’ cost structure says more than its explanation does. Marketing expense rose to 35.5% of revenue that quarter, up from 22.9% a year earlier, even as revenue itself shrank. A company spending a larger share of a smaller number to win customers is living the acquisition-cost problem, whatever the earnings release calls it. Store closures and distributor transitions may have driven the headline miss; the marketing ratio suggests the underlying pressure was there regardless.


The case against this reading deserves its strongest form. Amazon does not hand over the customer: it keeps the purchase data, owns the reorder, and prices the brand’s product beside Amazon’s own label, so a company that leans on the marketplace may be swapping a franchise for a queue it does not control. The marketplace is not even cheap in the way the story implies. Amazon’s cut across fees and advertising now tops half of many sellers’ revenue, a figure Marketplace Pulse traces to sellers now surrendering more than 50% of revenue to the platform. For the thesis to fail, the direct relationship would have to command a margin premium larger than Amazon’s acquisition savings, so that the holdouts are capitulating into a worse deal. That premium was always financed by cheap acquisition, and once the financing lapsed, the “owned” customer proved to be a customer leased from Meta on a rolling contract, with only one live question left: which landlord converts.

If paid acquisition stays expensive and Amazon keeps converting intent at a premium, “direct-to-consumer” reverts to what Modern Retail called it in 2019: a way to launch a brand, not a way to run one. Independence, it turns out, was a subsidy, paid for by a decade of cheap clicks that have stopped clearing. What the holdouts give up on the way to Amazon is the one asset the model existed to secure: the direct line to the buyer. Economic pressure is doing the persuading that no marketplace pitch ever managed. The boycott was never quite a principle. It was a budget.