Assaí added almost nothing to its store count and still tripled delivery sales. The Brazilian cash-and-carry chain reported last-mile sales up 204% across 104 stores in the third quarter of 2026, after a 237% jump through iFood in the second quarter, with gross revenue at R$21.3 billion and monthly traffic above 40 million visits. As of November 12, 2026, the interesting number for anyone running a store network is not the percentage: it is the fact that the growth came from turning existing aisles into picking lines rather than from new square metres.
A store base that did not move
Read the disclosures in order and the pattern is consistent. Gross revenue grew 2.7%, customer traffic grew 2.1%, and delivery sales grew by a multiple. The company is not opening its way into the channel; it is converting stores that already sit inside dense catchment areas into fulfilment points, then letting a third-party platform handle the last mile.
What it shows is category growth without shelf growth: the same floor space now serves a second demand channel, and consumer demand for delivery is being met from inventory that was already paid for. That is a different bet from the one quick-commerce startups made. They built dark stores sized for delivery only, with narrow assortments and no walk-in traffic. Assaí is running the opposite configuration: a large-format wholesale floor that also serves walk-in buyers, with delivery bolted on. The cost base is shared; so is the constraint.
Where the economics actually bind
Delivery from a large store carries a structural advantage and a structural tax. The advantage is assortment: a warehouse-club floor already holds thousands of SKUs, so the picking list rarely runs dry. The tax is picking cost, because a store laid out for browsing forces a picker to walk the same aisles a shopper would, at a fraction of the basket value.
The stock-accuracy problem
In a physical store, a missing item is a small inconvenience; the shopper picks something else. In delivery, the order has already been placed against a listing, so a missing item becomes a refund, a substitution call and a rating. Operators who scale delivery from stores usually discover that inventory accuracy, not rider speed, is the binding constraint: on-shelf availability, measured against the listing rather than against the shelf, is what decides whether a delivery order is filled whole.
Private label as the margin lever
Private label reached 140 SKUs in the quarter, with a year-end target of 250 across four tiers, and the company is expanding a supplements range into 300 stores. The delivery angle is straightforward: own-brand lines carry better margin, and margin is what pays for picking and the last mile. They do not solve availability, which depends on replenishment discipline at store level, and the distribution that feeds it is planned per store rather than per channel.
| Model | Cost base | Assortment | Main risk |
|---|---|---|---|
| Dark store only | Dedicated rent and labour | Narrow, fast-moving | Order density below break-even |
| Store as hub | Shared with walk-in trade | Wide, includes long tail | Picking cost and stock accuracy |
| Marketplace fulfilment | Paid per parcel | Extended range | Loss of control over service |
What other chains should copy, and what they should not
The copyable part is sequencing. Assaí ran the delivery channel through an existing platform first, learned what actually sold in small baskets, and only then began widening its own range. A marketplace pilot on Mercado Livre using a fulfilment model is the next step in the same direction: test demand before committing capital.
The part not to copy is the assumption that a wide assortment automatically wins. Quick-commerce operators argue the opposite: narrow ranges, high turnover and a small footprint produce better unit economics, because every shelf is a bet that pays back within days. Both models can work; they fail for different reasons, and the failure shows up in different numbers.
One caveat on the figures. The 204% and 237% figures come from company reporting on a channel that is still small relative to total revenue, and neither release discloses the absolute delivery revenue or the contribution margin. The 40 million monthly visits are a traffic count, not a buyer count. Treat the direction as reliable and the multiples as base-effect artefacts until the channel is broken out properly.
FAQ
Why does last-mile growth matter more than store growth?
The store base barely moved while delivery sales tripled, so the incremental revenue came from existing stores used as fulfilment nodes.
Is 204% comparable to a quick-commerce startup?
No. It is growth off a small base inside a large network. Compare contribution margin per order, not the percentage.
What breaks first when delivery scales?
Stock accuracy. Once an order is placed against a listing, a missing item becomes a refund and a bad rating.
Does private label help the delivery economics?
It helps margin, which pays for picking and the last mile. It does not fix availability.
