The Seller Blind Spot That Caps Every Marketplace's Growth

Based on the research ofMeng Li, Tao Li, and Lili Yu, "Retail Platform Analytics: Practice, Literature, and Future Research," Production and Operations Management, 2026

Li, Li, and Yu's review of retail platform analytics organizes the whole field into three themes: demand-side management, supply-side management, and matching the two. In principle, the three get equal billing. In practice, almost every visible piece of platform analytics infrastructure, the recommendation engine, the forecasting model, the personalization layer, lives on the demand side, where the metrics are legible, the wins are attributable, and the case studies write themselves. Supply-side analytics, the systems that would tell a platform which sellers are about to run out of stock, ship late, or defraud a customer, get built reactively, after a scandal, rather than as a parallel investment made with the same seriousness from day one. The review's own three-part structure is a quiet argument that this asymmetry is a choice, not a law of marketplace physics, and that the choice has a cost.

The Myth That Funded a Decade of Spending

The clearest evidence that buyer-side analytics attracts disproportionate attention is a statistic that turns out not to be true. For more than a decade, retail and marketing writing has repeated some version of the claim that 35 percent of what shoppers buy on Amazon comes from its recommendation engine. It traces to a single line in an October 2013 McKinsey Quarterly article, "How Retailers Can Keep Up With Consumers," which stated that "35 percent of what consumers purchase on Amazon ... come[s] from product recommendations based on such algorithms." Amazon has never confirmed the figure, in 2013 or since, and a 2026 statistics review found it still circulating in AI-marketing content thirteen years later, cited as though it were a live measurement rather than a thirteen-year-old unsourced estimate.

The irony is that the same 2013 McKinsey article buried a more consequential number a few paragraphs later: "Amazon ... acts as a traditional retailer in only 35 percent of its customer transactions. The majority of the products bought by Amazon's customers flow through its marketplace or its fulfillment services for third-party sellers." Even then, Amazon's growth engine was already majority seller-driven. The industry memorized the buyer-side statistic and built conference talks around it, while the seller-side statistic, the one saying most of the business ran through other people's inventory, disappeared into the same paragraph nobody quoted. One number about predicting buyers became legend; a more important number about managing sellers became a footnote.

What the Review Actually Argues

Li, Li, and Yu are not reporting a new experiment; their paper is a systematic synthesis of industry practice and academic literature on how retail platforms manage both sides of their marketplace. The abstract states the motivation plainly: retail platforms generate "a significant amount of customer and seller data that can be leveraged for advanced business analytics," yet "a systematic study of their operations is lacking in the literature." Their framework splits platform management into demand-side management, supply-side management, and matching, and for each theme they catalog current practice, review the relevant research, and propose open questions. The paper's contribution is the map, not a single finding, but the map itself makes the point: supply-side management, meaning how a platform understands, screens, and supports its sellers, is given equal structural weight to demand-side management, meaning how it understands buyers. The paper's quiet argument is that most platforms' actual practice does not match that structural equality, and that the mismatch is exactly the gap worth researching next.

The side you don't measure doesn't disappear. It becomes the ceiling you can't see until growth stops.

When the Seller Side Is a Black Box

The cost of under-measuring sellers is not hypothetical; it shows up directly in how two-sided marketplaces behave when buyers cannot verify seller quality for themselves. Liran Einav, Chiara Farronato, and Jonathan Levin's review of peer-to-peer markets for the Annual Review of Economics describes exactly this dependency: these markets "rely heavily on user feedback to provide ongoing monitoring" precisely because buyers cannot otherwise observe seller quality before transacting. When that monitoring is weak, the damage is measurable. Maryam Saeedi's structural analysis of eBay, published in the RAND Journal of Economics, found that removing the platform's reputation mechanism, its substitute for direct seller-side measurement, would increase low-quality sellers' market share, lower prices, and cut sellers' profit by 66 percent and consumer surplus by 35 percent. That is not a buyer-side problem being solved by a seller-side tool; it is proof that when a marketplace cannot see its sellers clearly, the entire market, both sides, gets worse.

Amazon's own practice shows both the scale of the problem and the lopsidedness of the response. Its Seller Central "Account Health" page is the closest thing it has to a seller-side analytics product, an overview of "adherence to the policies and performance targets required to sell in the Amazon store." It is a compliance scorecard, built to flag and remove violators, not an investment-grade layer built to help good sellers forecast demand or manage inventory risk the way the buyer-facing recommendation engine helps shoppers. Where Amazon invests at real scale on the seller side, it does so defensively: its 2024 Brand Protection Report discloses that the company "invested more than a billion dollars and employed thousands of people," including machine learning scientists and investigators, to fight counterfeiting and fraud, and that its proactive controls now block more than 99 percent of suspected infringing listings before a brand ever has to report them. That is genuine, well-funded supply-side analytics, but it exists to catch the worst sellers, not to grow the median one. Jungle Scout's survey of nearly 4,900 sellers found that 20 percent said their own accounts had been suspended by Amazon at some point, a governance signal, not a growth one. The seller side gets investment once it threatens to become a liability. It rarely gets invested in as a growth lever on its own.

The Ceiling Is Already Visible in the Seller Count

If the under-measured side is really what caps marketplace performance, the clearest evidence should be structural: a shrinking or struggling seller base even as that base drives most of the business. Marketplace Pulse data shows exactly that pattern on Amazon. Active seller count on Amazon.com fell from 584,000 in January 2025 to 500,000 by March 2026, even as third-party sellers account for 62 percent of units sold and roughly 69 percent of total marketplace GMV, and even as 47 percent of sellers reported a year-over-year margin decline. New seller launches told the same story from a different angle: a separate Marketplace Pulse analysis found just 165,000 new sellers listed their first product on Amazon.com in 2025, the lowest figure the firm has tracked since 2015 and down 44 percent from 2024. A platform whose own demand-side engine, recommendations, forecasting, personalized search, keeps improving every year cannot keep growing if the supply it is matching buyers to is contracting and souring underneath it. The buyer-side dashboards can show green for years while this is happening, because they were never built to measure the thing that is actually breaking.

This is the practical version of what Li, Li, and Yu's review is arguing in academic language: a platform's competitive edge rests on converting two-sided data, buyer and seller, into operational and matching decisions, not on how much data it accumulates about either side alone. A platform that measures buyers exhaustively and sellers only when they go wrong is not under-measuring at random. It is systematically blind to the input that, per the paper's own supply-side and matching themes, determines whether the demand-side engine has anything good to match buyers to. The fix is not to spend less on personalization. It is to fund seller-side analytics, inventory risk, fulfillment reliability, proactive health scoring, fraud detection, as a parallel growth investment rather than a cleanup budget, and to ask, the next time a demand forecast beats expectations, whether the supply behind it can actually hold.