SHEIN stopped trying to predict which clothes would sell and just asked the market directly, 100 units at a time

the problem

Fashion demand is genuinely unpredictable months in advance, so any forecast-then-mass-produce model is betting real capital on a guess — and the guess is wrong often enough that unsold inventory becomes the industry's largest structural cost

background

what everyone would do

The standard fix for a forecast that's often wrong is a better forecast — hire sharper trend forecasters, run bigger focus groups, mine more historical sales data, invest in demand-prediction software. That treats the error as a modeling problem to be reduced, but months-ahead fashion demand has an irreducible unpredictability no model closes entirely, so even a meaningfully better forecast still commits large capital to guesses that are wrong often enough to generate the industry's chronic unsold-inventory writeoffs.

what they saw

SHEIN's operators saw that the actual bottleneck wasn't forecast accuracy at all, it was batch size: traditional manufacturing's minimum order quantities were large enough that testing a design in the real market was as expensive and risky as committing to it, so retailers had to predict first and could only afford to test a handful of styles per season. If a supplier could be persuaded to accept an order of just 100-200 units, the cost of finding out whether a design actually sells collapses far enough that the retailer no longer needs to guess in advance — it can just run the real transaction as the test, for far more designs than any forecast-driven competitor could afford to gamble on.

the move

Because a 100-unit test batch is cheap enough to run on far more designs than a traditional large-batch model could ever risk, SHEIN can test roughly five times as many distinct styles per unit of capital committed as a competitor testing the same total volume in fewer, larger batches — replacing forecasting accuracy with test-batch volume, since running many cheap real-world tests finds more genuine hits than trying to predict hits in advance ever can.

why it works

By shrinking the minimum viable test to 100-200 units, SHEIN converts every new design into a cheap real-world experiment rather than a forecasted bet: actual purchase data, not a prediction, decides within days whether a style gets reordered at scale or dropped, which is a strictly more reliable signal than any forecast because it is the actual behavior being predicted, not a proxy for it. Because each individual test is cheap, SHEIN can run vastly more of them per unit of capital than a competitor placing fewer, larger forecasted orders, so it finds more genuine hits by trying more designs rather than by guessing better which few designs to try — which is why it can list thousands of new styles daily while keeping inventory-writeoff rates far below the industry's traditional levels. The mechanism only holds together because SHEIN also built the supplier density and payment terms (fast settlement, geographic concentration in Guangzhou's manufacturing clusters) that make hundreds of small, fast-turnaround orders logistically and financially viable for the factories accepting them, not just cheap in theory.

the payoff

SHEIN scaled to an estimated $23 billion in 2022 sales and roughly $32.5 billion in 2023, adding between 2,000 and 10,000 new product listings daily at peak — a pace of style-testing volume unmatched by traditional retailers using forecast-driven production — while its valuation rose from roughly $5 billion in 2019 to as high as $100 billion in 2022 on the strength of the model's efficiency.

where it breaks

This mechanism depends on suppliers being willing and able to accept orders far smaller than their standard minimum, which usually requires either a geographically concentrated, flexible manufacturing base (as SHEIN found in Guangzhou) or enough buying power and fast payment terms to make small orders worth a factory's time; a retailer without that leverage or that supply base simply can't get the low minimum order quantities the model depends on. It also depends on the product category having a short enough production and shipping cycle that a real-world test can return a usable signal before the trend it's testing has already passed — a category with long lead times (heavy manufacturing, anything requiring tooling or certification) can't convert to cheap iterative testing the way a T-shirt can. And running many more small tests than a forecast-driven competitor only pays off if a large fraction of the untested designs would otherwise have been guessed wrong; in a market where demand is genuinely easy to forecast, the extra testing volume is pure overhead with no offsetting advantage.

what came after

小单快返 is now studied extensively in Chinese supply-chain and e-commerce literature as SHEIN's core structural advantage over both traditional fast fashion (Zara, H&M) and other online apparel sellers, and Chinese manufacturing clusters around Guangzhou have reorganized specifically to support small-batch, fast-turnaround orders at the volume SHEIN's model requires.

references

1. [1]澎湃新闻 (The Paper) — 复刻SHEIN，中国跨境供应链大突围The Paper (澎湃新闻), 2023thepaper.cn
2. [2]SHEIN Is Meeting Customers' Demands Through Reimagining the Supply ChainWWD (Women's Wear Daily), 2023wwd.com

keep it

same kind of clever

• 1958 · Bank of America (the 'Fresno Drop')No bank knew if ordinary people would actually use unsecured credit — so instead of running a pilot with volunteers, Bank of America just mailed 60,000 live credit cards to an entire city and watched what happened.
• 1930 · Starrett Brothers and Eken / Empire State BuildingA skyscraper deadline was too tight to finish the design, then demolish the old building, then build — so the contractors stopped waiting for each phase to end and ran all three at once.
• 240 BCE · Eratosthenes, Library of AlexandriaEratosthenes measured the size of the entire planet without ever leaving Alexandria