4 hours, 57 billion reais and 1,300 stores are the three numbers that now define a delivery promise in Brazil, and as of November 13, 2026 they are quoted in the same sentence by almost every large retailer. Shopee runs Turbo with a four-hour promise across Greater Sao Paulo, Mercado Livre delivers on the same day including Sundays, Amazon launched a fifteen-minute essentials service on top of Rappi logistics, and Magalu is wiring more than 1,300 stores as support points for one-hour delivery. Behind those promises sits a quieter change: in O2O terms the store is no longer only a place where a shopper walks in, it is a node that has to hold stock, absorb picking labour and hand parcels to a courier on a schedule.
What a four-hour promise actually requires
The four-hour number is a promise about geography, not about effort. A node can only ship within the window if the item is already inside the radius, so every hour removed from the promise forces stock one step closer to the buyer. That is why the same operators that advertise hours are also building square metres: capacity and proximity are the same investment counted two ways.
For a brand selling through those networks, the practical question changes. It is no longer whether the product is listed in a city but whether it is present in the specific node that serves the postcode where the order lands. Distribution coverage measured by city hides the gap that a customer experiences in one neighbourhood.
Three layers, three failure modes
Pickup is not fulfillment
A store used as a pickup point needs a counter and a shelf. A store used as a dispatch node needs a picker, a packing bench, a staging area and a handover slot that does not collide with the checkout queue. Operators routinely upgrade the first without funding the second, and the symptom is a delivery promise that holds in the morning and collapses after four in the afternoon.
Labour is the cost that does not appear in the lease
Converting a store into a node does not require new rent, which is exactly why the decision looks cheap on a spreadsheet. The hours come from the same pool that serves shoppers in the aisle. When volume through the store rises without adding hours, the queue simply moves from the checkout to the pick list, and the second queue is invisible to the customer until the parcel is late.
| Node role | What it must hold | Failure mode | Metric |
|---|---|---|---|
| Dedicated fulfilment centre | Depth in fast movers | Stock in the wrong region | Fill rate by node |
| Cross-dock point | Almost nothing overnight | Cut-off slips in peak | On-time handover |
| Store as dispatch node | Top sellers plus local assortment | Picking competes with shoppers | Pick time and queue |
| Store as pickup point | Reserved parcels only | Space taken by uncollected orders | Dwell time per parcel |
The table is a sorting tool, not a maturity ladder. Most Brazilian networks now run all four roles at once, which is why a single delivery metric cannot explain performance. A brand that asks for one number will get the average of four different operations.
Density redraws the trade area
A store that ships competes in a smaller trade area than a store that only sells. The catchment shrinks from the distance a shopper will walk to the distance a courier can reach inside the promise, and the two are not the same shape. Stores that looked marginal on footfall can become the best nodes in the network, while a high-traffic flagship with no back room can be the worst. That inversion is why store selection for delivery cannot reuse the ranking built for retail traffic.
The screen has to be rebuilt from three inputs: order density by postcode, the time a pick actually takes in that store, and whether the assortment that sells online is on the shelf. A high-potential store under this lens is one where all three line up, and there are usually far fewer of them than the store count suggests. Naming those stores is the useful output of the exercise, because investment follows names, not averages.
Four checks before the peak weeks
The checks below are cheap and they fail loudly, which is what a peak season needs. None of them require a new system. Each one produces a list with store codes attached, and a list with codes on it is the only output that survives the next planning meeting.
FAQ
Does same-day delivery pay for itself in Brazil?
Not through delivery fees. Consumers rarely accept paying more for speed, so the return comes from conversion and repeat orders. Treat the promise as a store selection question and the economics become readable: nodes that hold the online assortment convert, nodes that do not simply absorb cost.
Is a dark store always better than a retail store?
No. A dark store controls labour and layout but cannot use existing inventory. A retail store carries stock already paid for but shares labour with shoppers. The right answer depends on order density in the trade area, which is why a mixed network usually beats a single format.
How should distribution coverage be reported to management?
By node and by item for the fastest sellers, not by city. City-level coverage hides the case where the item is listed in the market but absent from the node that serves the order, and that gap is what the customer actually experiences.
