As of November 10, 2026, the two most quoted numbers in on-demand retail point in opposite directions. Coupang reported third-quarter revenue of US$9.96 billion, up 7 per cent year on year, against an operating loss of US$450 million and a net loss of US$501 million, its third consecutive quarterly loss. In the same period BIG W placed 35,000 products from 156 stores on DoorDash, promising delivery in as fast as 60 minutes to more than 80 per cent of Australians. Revenue growth and losses are arriving together, and the line that explains both is capacity.
Two numbers, one bottleneck
Read as an industry trend, the pattern is familiar from every delivery build-out of the past decade. The first wave buys reach: more stores, more postcodes, more promises. The second wave discovers that reach is not the constraint, capacity is. A store that can assemble forty orders an hour and receives eighty does not grow at twice the rate, it fails at twice the rate, and the cost shows up as refunds, credits and lost repeat orders.
The consumer trend underneath is what makes this harder to price. Shoppers are not buying more, they are buying later. Australian retail turnover reached $39.91 billion in the year to September, but the growth in discretionary categories was thin, and food turnover rose 3.3 per cent against food price inflation of 3.2 per cent. That is a market where volume is flat and the willingness to pay for speed is concentrated in a narrow set of missions.
Capacity is bought in shifts, not in vans
The picking hour is the real unit
Every fulfilment promise converts into a picking hour. A 60-minute promise on 35,000 products means a store must be able to find and assemble an order inside a window that also serves walk-in customers. When the promise tightens, the cost does not rise linearly; it rises in steps, because the store has to hold labour in reserve for the worst hour rather than the average one.
Batching decides whether the step is affordable
Batching is the only lever that reduces cost per drop without lengthening the promise. Two orders picked in one walk and delivered on one route cost less than two separate trips. But batching needs order density inside a short window, which is exactly what a thin trade area cannot supply. This is why the same operating model produces healthy margins in one suburb and losses two postcodes away.
DoorDash now lists more than 55,000 grocery, convenience and retail stores across its markets, and the marginal store added is almost always in a thinner catchment than the last one. Growth in store count therefore dilutes order density before it improves it, which is the mechanism behind a widening operating loss on rising revenue.
What store-level availability does to the promise
Availability is the second half of capacity. A store can have labour and still miss, because the item is in the back room, on a pallet, or in a tote that nobody has unpacked. BIG W is explicit that stock is subject to local store stock, which is an honest description of the constraint: the promise is national, the inventory is per store.
| Constraint | Signal to watch | Threshold that hurts |
|---|---|---|
| Picking labour | Orders per staffed hour | Above 85 per cent utilisation |
| Order density | Drops per route hour | Below two drops |
| Store availability | Lines substituted per order | Above one in ten |
| Promise integrity | Orders late against the slot | Above 4 per cent |
The table is deliberately narrow. Teams tend to track revenue per store and delivery cost per order, both of which lag the problem by a quarter. Utilisation, density, substitution and lateness move within days, which is the only timescale that matters when the promise is measured in minutes.
The four numbers to hold before the peak
Before the busiest trading weeks, the useful work is not adding stores but testing whether existing ones can hold the promise under load. Run one store through a controlled peak, record where the order stops moving, and fix that point before the volume arrives.
FAQ
Does a widening operating loss mean on-demand delivery does not work?
No. It means capacity was bought ahead of density. The model works where order density per route hour is high enough to batch, and loses where it is not.
Why focus on picking hours rather than delivery cost?
Because the promise is set by the store, not the road. A late order is almost always a late pick, and the cost of protecting the pick is what shows up in the operating line.
Is store stock a fair excuse for a missed promise?
It is an explanation, not an excuse. If availability is per store, the promise should be qualified per store, or substitutions should be offered before checkout.
What single metric predicts margin best?
Drops per route hour. It captures density, batching and catchment quality at once, and it moves fast enough to act on.
