1.02 trillion yuan, 2.4 percent, and tens of billions in delivery subsidies: three numbers that describe the same quarter. The first is gross merchandise value across major platforms in the three weeks to the end of October, the second is its growth, and the third is what Alibaba, JD.com and Meituan have committed to one-hour delivery, as of November 11, 2026. Put together, they point at one reallocation, from coupon budgets to distribution capacity, and at a new test for brands: not how many stores are listed, but how many can complete an order inside the promised window.
Distribution nodes replace coupon budgets
Subsidies are easy to measure and easy to stop. Capacity is neither. Once a platform pays for a warehouse, a pick line or a store retrofit, the cost is fixed and the payback depends on volume per node. That arithmetic is what pushed the three platforms toward different answers in the same period, and it is why the next round of competition will be decided on coverage rather than on discount depth.
The visible effect is a quieter kind of growth. Order volume can rise while average discount falls, as long as each node ships more and each shipment is cheaper to assemble. That combination only works when the catalogue is trimmed to what a node can actually hold, which is a merchandising decision rather than a marketing one.
What a node has to do that a store did not
Pick density beats floor space
A store is laid out for a person walking with a trolley. A node is laid out for a picker walking against a timer. The same square metre can hold twice the sellable range if the fast movers sit within arm's reach of the packing table, and half the range if the layout follows the shopping path instead. Retrofitting a store into a node is therefore a question of metres, not of investment size.
The reservation problem
Inventory sitting in a store is already promised to whoever walks in next. Once the same unit is also listed online, two demand streams compete for one physical item, and the loser is whoever arrives second. Platforms solve this with reservation windows and buffer stock, but both cost coverage. A brand that ignores the buffer will see phantom availability: the system says yes and the shelf says no.
Three node types, three failure modes
Node types fail differently, which is why a single fill-rate target hides more than it reveals. A dedicated warehouse fails on assortment breadth, a converted store fails on pick speed, and a partner outlet fails on willingness to prioritise an online order over a walk-in customer.
| Node type | Range it can hold | Typical failure | First fix |
|---|---|---|---|
| Dedicated fulfilment warehouse | Narrow, deep in fast movers | Long-tail SKUs unavailable | Cross-node transfers on a fixed schedule |
| Converted store | Wide, shallow | Pick time collapses at peak | Front-of-store pick zone and a fixed route |
| Partner outlet | Wide, inconsistent | Online orders deprioritised | Separate incentive tied to completion rate |
The table also explains why coverage numbers reported at platform level rarely match what a brand sees in its own data. A store counted as connected may be listed but not stocked, or stocked but not staffed during the evening peak. Coverage that is measured by connection rather than by completion is a vanity metric.
A quarterly routine for node coverage
None of this requires new software. It requires agreeing on one definition of coverage and applying it to every node type, including the ones that were never designed to fulfil orders. Brands that skip that step discover the gap during a peak, when there is no time to fix it.
FAQ
Is coverage the same as distribution?
No. Distribution asks whether a store carries the product. Coverage asks whether that store can complete an order for it within the promised window, which also depends on listing status, staffing and pick capacity.
Why do platforms report node counts instead of completion rates?
Node counts are stable and easy to audit; completion rates move daily and depend on how each order is timed. The first is a capital story, the second is an operating story, and only the second predicts whether a peak will be served.
Does this apply to categories outside grocery?
It applies wherever the promise is measured in hours. It matters least for categories bought on a planned basis, where a two-day delivery is acceptable and the store network is irrelevant.
What should a brand do first if it has no node data?
Start with the top fifty SKUs by revenue and ask one question per store: can this store hand over this item today, at eight in the evening. The answer set is small enough to collect manually and specific enough to act on.
