A network can double its store count and still lose sales to an empty shelf. Amazon Now crossed $1 billion in annualised gross sales while running more than 750 micro and urban fulfilment centres across 60-plus Indian cities, according to Zacks. As of November 15, 2026, the harder question is not how many cities are covered but whether each node keeps a fast mover pickable when an order lands. Coverage is built with capital; availability is built with replenishment discipline, and the two rarely scale at the same speed.
Coverage is a lagging indicator
City counts tell a brand where the network was built. They say nothing about whether a specific SKU is on the shelf at 9pm on a Tuesday. A tier-2 node handling a few hundred orders a day cannot economically hold metro-level depth, so its assortment gets cut, and the cut items are exactly where availability breaks first. For a brand, being listed in 60 cities and being available in 60 cities are different claims that often get reported as one.
The festive numbers make the gap visible. Quick commerce took about 16% of online festive sales this year, up from 12%, and grocery and gifting moved disproportionately onto the channel. Order volumes rose 75% while GMV rose about 60%, which means baskets got smaller. Smaller baskets spread the same pick, pack and delivery cost over fewer items, so every stockout costs proportionally more than it did a year ago.
| Metric | What it shows | Why it matters |
|---|---|---|
| City count | Coverage breadth | Lagging: where the network exists |
| Node fill rate | Item availability | Leading: whether stock is pickable |
| Order frequency | Demand density | Decides if node economics hold |
| Basket size | Items per order | Smaller baskets raise cost per stockout |
Where availability actually breaks
Availability fails in three places, and each has a different owner. The first is listing: a product that is not correctly uploaded, categorised or attributed never enters the candidate set at all. The second is inventory accuracy: the storefront says in stock while the shelf is empty, which is worse than not listing, because the cancellation carries a platform penalty. The third is time-of-day: night shifts staffed to walk-in traffic rather than online orders cut availability precisely when a dark store has its clearest advantage over a supermarket.
Most teams report one number, in-stock rate, and then argue about whether it is good. Splitting it into listing coverage, inventory accuracy and time-sliced availability makes the argument productive, because the fixes live in different departments: catalogue ops, warehouse, and workforce scheduling.
Sample by node, not by region
Regional averages hide the problem they are supposed to reveal. A state-level view of 94% availability can sit on top of forty nodes at 99% and ten nodes at 60%, and the ten are where the growth is supposed to come from. Pull the sample at node level, keep the top twenty sellers fixed, and refresh the panel as the footprint moves into new towns.
Replenishment rules for thin-density nodes
Thin nodes need a different rule set, not a smaller version of the metro rule. Order frequency, not order count, decides whether a node can justify depth. Where frequency is low, the practical answer is a narrower core assortment replenished more often, with long-tail items available on a longer promise rather than shown as in stock and cancelled later.
Assortment cuts should follow demand data rather than category logic. A node that sells out of one SKU three weeks running while another sits untouched is telling you which cut is safe and which is not, and that signal only exists if availability is logged per item per day rather than averaged into a monthly figure.
Log the reason, not just the rate
A stockout rate without a reason code cannot be acted on. The five codes worth keeping are: not listed, listed but out of stock, listed but mis-categorised, out of stock during a specific time window, and cancelled after order. Each maps to a different team and a different fix, and the distribution across codes is usually stable enough to show whether a change worked within two weeks.
The expansion story is real, and the $1 billion milestone does mark a shift from experiment to infrastructure. But the brands that hold share in these networks will be the ones that treat availability as a per-node, per-item, per-hour metric rather than a headline about how many cities were added.
FAQ
Why does fast network growth raise stockout risk?
Each new node carries less inventory depth and serves a thinner demand base, so assortment has to be cut, and availability breaks first on the items that were cut.
Is city count a useful availability metric?
No. It shows where the network was built, not whether stock is pickable. Item-level fill rate inside nodes is the leading indicator.
How should a brand sample this network?
At node level, on a fixed panel of top sellers, refreshed as the footprint moves. Regional averages hide the nodes that matter most.
What does a falling basket size change?
It raises the cost of every stockout, because pick, pack and delivery costs are spread over fewer items per order.
