The shopping festival that stopped publishing its score

For eighteen years this date arrived with a counter that never stopped climbing. This year the most conspicuous thing in the reports was the absence of one. No record banners, no hourly refresh, no triumphant total. A festival built on announcing its own size suddenly decided that size was no longer the point.

Start with the arithmetic. Third-party tracking put first-wave marketplace volume at 1.0263 trillion yuan, up 2.4% year on year; a year earlier the same window had grown 14.2%. Analysts put the full cycle somewhere between 3% and 6%. Retail sales overall rose 1.1% in the first three quarters and online retail 4.4%. None of that describes a single platform losing; it describes a curve changing slope.

The rules moved more visibly than the totals. Long-running cross-store thresholds were retired in favour of official instant discounts. Deposit pre-sales gave way to in-stock shipping with thirty-day price protection. Short-video platforms pushed straight per-item cuts. The discount window actually stretched to a month, but orders spread across it instead of piling onto one night, and the cost of deciding went down.

There is an unflattering logic behind that subtraction. Once shoppers stop assuming this is the cheapest day of the year, an elaborate coupon puzzle stops being a conversion tool and becomes a conversion obstacle. When new-user growth flattens, the contest shifts from acquiring people to letting regulars hesitate one less time. Simplification here is not generosity; it is friction removal.

The more interesting addition is the shape of the shelf. Services entered the main transaction path for the first time rather than hanging off a physical order as an afterthought. Cleaning, home maintenance and installation were broken into orderable units, producing 150 million service orders in the first half alone, with domestic services up more than 240%. Furniture and bathroom goods began bundling delivery, unboxing and fitting into the transaction itself.

Then came goods with no physical form. Model tokens and compute packages were listed as standard items for the first time, and one livestream sold nearly 100,000 membership cards in half an hour, representing roughly two trillion tokens of usage. From discounted goods to services to raw compute: nobody drawing the original plan in 2009 could have sketched that line.

On the shopper side, assistants looked like a new front door had opened. One passed 300 million monthly active users; another reported close to 80 million. A survey found nearly 80% of buyers consult an AI before ordering, and AI influence on purchase decisions edged past conventional advertising for the first time, 43.7% against 42.6%. On paper, the entry point had already changed hands.

Then the retention numbers arrive: under 15% at seven days, about 3% at thirty. Trying something once and building a habit are an order of magnitude apart. Most people treat the assistant as a novel button available during a sale, not as the place every purchase begins. A large curiosity spike is not an installed position.

The harder constraint is that every assistant sees only its own catalogue. A helper that can search one warehouse cannot answer the plainest question a shopper has, which is whether the same thing is cheaper somewhere else entirely. Smart comparison today mostly means faster search inside a walled garden, and taking the wall down is precisely the step no platform is eager to take.

So my reading is that this is an enhancement layer between the search box and the shelf, not a new transaction gateway. It saves the time spent flipping pages and decoding discounts. It cannot absorb the risk of choosing badly, overpaying, or failing to return an item. While that risk stays with the person, the gateway stays where it was.

The merchant side is where the change is already concrete. Single-purpose tools for copy and customer chat are being stitched into a continuous stack spanning selection, listing, bidding, inventory routing and after-sales. What it compresses is operating cost, not missing demand. Mistaking a cost tool for a revenue engine is the easiest error of this cycle.

Platforms stopped publishing final sales figures and now report active buyers, repeat rates and retention instead. That is a sign of maturity, and also a small trap: when a slower growth rate is uniformly re-labelled as higher-quality growth, the metric changes but the question does not disappear. An efficiency race is only won if shoppers genuinely waste less money.

Eighteen years on, a loud promotion has turned into an annual stress test of goods, fulfilment, services and information agents. The page with no headline number is the one that finally shows the whole exam paper: whoever makes choosing simpler earns the next repeat order.