Store-based quick commerce

Quick commerce spent its first decade assuming that fast delivery required a new building. The public numbers no longer support that assumption. This page sets out the alternative — fulfilling rapid grocery orders from supermarkets that already exist — and tests it against what listed operators actually report.

A working definition

Store-based quick commerce is rapid online grocery delivery fulfilled from a trading supermarket rather than from a purpose-built dark store. Pickers move through the live shop floor alongside walk-in customers. Orders are staged at a dispatch point near the front of the building. Riders collect from the same site that has been serving that neighbourhood for years.

It is sometimes called store picking, store-fulfilled delivery, or in-store fulfilment. The vocabulary matters less than the structural fact underneath it: no new lease is signed to serve the online customer. The delivery business rides on retail infrastructure that another P&L already justifies.

That single fact drives almost everything else on this page.

Where the industry actually stands

The honest starting point is that pure-play quick commerce, after roughly six years and enormous capital, has produced exactly one operator reporting a positive margin — and it is a very thin one.

Operator	Period	Sites	Reported margin
Blinkit (Eternal)	Quarter to Jun 2026	2,443	Adjusted EBITDA +0.6% of NOV1
Swiggy Instamart	Quarter to Mar 2026	1,143	Contribution margin −1.8% of GOV (March exit −1.1%)2
Zepto	Year to Mar 2026	1,139	Net loss INR 5,905 crore3

Set that against the capital consumed getting here. Getir raised roughly $1.8bn and peaked at an $11.8bn valuation in March 2022.⁴ It exited France, Italy, Spain and Portugal in 2023; the US, UK, Germany and the Netherlands in 2024; and shut down Gorillas, which it had bought for €1.1bn barely a year earlier. In February 2026 Uber acquired its Turkish food-delivery arm for $335m and took 15% of the remaining grocery business for $100m.⁵

From $11.8bn to a $435m transaction in under four years. The product was never the problem — customers liked fifteen-minute groceries and still do. The lease was the problem.

The $2.6bn natural experiment

Arguments about fulfilment models are usually untestable, because no operator runs both at scale and publishes the difference. In 2025 one did, by accident.

Kroger had spent years building Ocado-powered automated customer fulfilment centres — the most capital-intensive version of the build-a-new-box strategy. In November 2025 it announced the closure of three of them, taking $2.6 billion in impairment and related charges, and said it expected roughly $400 million of profitability improvement in 2026 from the shift. The replacement strategy was explicitly store-based fulfilment, third-party marketplaces, and what it called capital-light, store-based automation.⁶

The follow-up is the part that matters. In the quarter to May 2026, Kroger reported that its entire e-commerce business had turned profitable — ahead of schedule — with digital sales up 19% year on year.⁷

In markets where we have a store presence, we retained nearly all of those households and successfully converted them to store-based delivery and pickup. David Kennerley, CFO, Kroger — Q1 FY2026 results

Read that sentence carefully. The customers did not care which building their groceries came from. They cared about getting them. The automated warehouse was solving a problem the customer did not have, at a cost the business could not carry.

The other side of the same transaction shows up in Ocado's own accounts, where H1 2026 results are so distorted by partner-closure income that the company reports adjusted EBITDA both including (£432m) and excluding (£81m) it.⁸ Two independent sets of audited numbers describing one decision.

Store-based vs dark store

The two models are often discussed as if one is simply a cheaper version of the other. They are not. They have different cost shapes, different failure modes, and different natural habitats.

	Store-based	Dark store
Site cost	Already carried by in-store trade	New lease and fit-out from day one
Inventory risk	Shared with walk-in demand; stock turns daily	Carried alone; slow lines die on the shelf
Catchment proof	Demonstrated by years of footfall	Modelled, then discovered
Assortment	Wide — the whole shop is available	Narrow by design, typically a curated core
Pick efficiency	Lower — layout serves shoppers, not pickers	Higher — layout is engineered for picking
Speed ceiling	Constrained by floor congestion at peak	Constrained only by dispatch and distance
Time to launch	Weeks	Months
Downside if wrong	Some wasted labour; you stop	A lease you cannot exit

Read that table honestly and the trade is clear. The dark store is the better machine. The store is the better bet. A purpose-built site will out-pick a supermarket aisle every time; it simply has to carry a fixed cost that the supermarket does not.

It is the same trade a shipping line makes between a purpose-built container terminal and a general cargo berth. The terminal moves boxes faster. The berth was already paid for.

Why the cost structure differs

Cost per order in rapid grocery decomposes into roughly four things: the site, the pick, the drop, and the waste. Store-based and dark-store models sit differently on each.

The site

A dark store's rent and fit-out are fixed costs allocated across whatever volume it achieves. Early in life, when volume is thin, that allocation is brutal — and it is precisely when a new site has the fewest orders to spread it over. Eternal now assumes roughly INR 2.5 crore of capex per dark store, raised from INR 1 crore, against a steady-state assumption of INR 11 lakh of net order value per store per day.¹ Those two numbers together are the whole build-a-box case: a site has to reach and hold a demanding daily run rate before the capital behind it makes sense.

The store-based model does not carry this line in any meaningful sense, because the building's economics were settled by in-store trade long before the first online order arrived.

This is the whole argument, and it is worth being blunt about why it mattered so much. When capital was free, the fixed-cost hole was easy to ignore. When rates rose, it became the thing that ended companies.

The pick

Here the dark store wins, and it is not close. A site laid out for picking — short paths, dense faces, no trolleys in the way — will beat an aisle-picked supermarket on items per hour. Anyone claiming otherwise has not stood in a store at six in the evening.

The store-based model claws this back three ways: by keeping the online assortment tight rather than exposing the full range, by designing pick paths against real order composition instead of store layout, and by scheduling picking labour against the store's own trading curve rather than a flat roster.

It is worth noting how much headroom exists here even in the most automated operations. Ocado Retail reported units per hour up 11% and customer fulfilment centre costs at 5.7% of sales in H1 2026⁸ — meaningful improvement, in a purpose-built robotic warehouse, a decade into the programme. Picking productivity is a long grind wherever you do it.

The drop

Roughly a wash, with an edge to whichever asset sits closer to the customer. In dense urban markets that is frequently the supermarket, because supermarkets were sited where people live decades before quick commerce existed. Delivery cost is mostly a function of distance and rider utilisation, and a shorter radius is worth more than a faster pick.

The waste

The quiet one, and the one that decides more cases than people expect. A dark store's inventory is exposed only to online demand. If a line stops selling online, it stops selling.

The scale of this is visible in the disclosures. Blinkit reported inventory losses of 1.8% of net order value in the quarter to June 2026 — approximately INR 308 crore in a single quarter.¹ Hold that number next to the 0.6% adjusted EBITDA margin in the same quarter. Shrink alone is three times the profit margin. In a store, walk-in customers keep stock turning every day, and fresh categories — where the risk concentrates — are managed by an operation that has been doing it for years.

The densification trap

There is a second-order effect that rarely makes it into strategy decks, and the Indian market has just produced clean evidence of it.

Between April and July 2026 the top five Indian quick-commerce players added roughly 900 dark stores. Blinkit went to 2,511, Flipkart Minutes past 1,000, Amazon Now to 600–700, Zepto to 1,345, Instamart flat at 1,187 — an industry total somewhere around 6,650 to 6,750 sites.⁹ Expansion, on the face of it.

But those 900 stores entered only 152 new pin codes. Competitive overlap in metro pin codes rose from 26% to 44% in the same window.

That is not expansion. That is four operators building boxes on top of each other in catchments that were already served — splitting the same demand across more fixed cost. Every one of those leases has to be paid whether or not the order lands with you.

It is a familiar pattern to anyone who has watched a petrol-station price war, or four coffee chains opening on the same corner. The first site in a catchment earns a return. The fourth earns a lease obligation. And the operator who already owns a shop in that catchment is the only one who did not have to sign anything to compete.

What is genuinely hard about it

Advocacy without the difficulties is marketing. Store-based quick commerce has four real problems, and pretending otherwise is how conversions fail.

Co-existence. The online order and the customer in aisle seven sometimes want the same last pack of chicken. Someone has to lose, and the rules for who loses need to be decided in advance rather than improvised by a picker under time pressure. This problem simply does not exist in a dark store.

Congestion at peak. Online demand peaks when the store is busiest. Pickers and shoppers compete for the same physical space at exactly the wrong moment. Managing this is a scheduling and layout problem, not an attitude problem, and it is the single most common reason a promise window slips.

Stock accuracy. A shelf that shows two units may have none, because someone put one in a trolley four minutes ago. Substitution logic that customers tolerate — and that does not quietly destroy basket value — is harder to build than it sounds.

Organisational ownership. The store manager is measured on store performance. Online fulfilment consumes their labour and their stock, and initially credits neither. If the operating model does not resolve this, the model fails for reasons that have nothing to do with logistics.

Every one of these is solvable. None of them is solved by enthusiasm.

How a conversion actually works

Converting a trading store into a fulfilment node is a sequence, and skipping steps is where most programmes lose their economics.

1. Cut the assortment before anything else. Not the full range. The lines that actually get ordered for immediate delivery, which is a far narrower set than merchandising instinct suggests. Every additional line lengthens the pick path for every order, including the ones that do not contain it.
2. Design the pick path against real orders. Sequence the online assortment by how baskets are genuinely composed, not by how the store is laid out for browsing. This is where most of the recoverable minutes live.
3. Fix a dispatch point. A defined staging area near the front, with a defined handover to riders. Orders that wander around looking for their rider are orders that miss their window.
4. Schedule against the store's own curve. Picking labour needs to flex with both online demand and floor congestion — which peak together, and are not the same constraint.
5. Instrument the promise, not the average. Average delivery time hides everything that matters. Measure the tail: the proportion of orders that miss, and why. Averages improve while customer experience degrades, and you will not see it.
6. Give the store a stake. Online volume must show up in how the site is measured and rewarded. Otherwise it is a tax, and it will be treated as one.

What breaks at scale

Plenty of fulfilment networks look excellent at ten sites and come apart at sixty. The reason is nearly always the same: the pilot ran on the goodwill of a few strong store managers, and goodwill does not replicate.

At scale the constraints change character. Assortment discipline that one motivated site maintained by hand has to become a system. Pick-path design has to survive stores with different footprints. Rider dispatch stops being a per-store problem and becomes a network one, because riders can be shared between nearby sites and utilisation is where the last real cost reduction lives.

Tesco's Whoosh is the closest thing to a proof that this scales: roughly 1,800 stores, reach across more than 70% of the UK population, sales up 51% year on year, and rapid delivery now accounting for 14% of Tesco's online FMCG sales.¹⁰ No new fulfilment estate was built to get there.

A really meaningful business … quite a long way to go before we would say that model is mature. Ken Murphy, CEO, Tesco — on Whoosh, July 2026

The second scale problem is multi-channel. Once a single store estate serves own-brand rapid delivery, a scheduled online supermarket, and third-party marketplace partners, three different promise windows compete for the same inventory and the same pickers. That is the hardest operating problem in this field, and it is where operating models actually break — not at launch, but at the point where a second and third channel arrive on top of infrastructure designed for one.

What this means in the Gulf

The GCC is an unusually good market for the store-based model, for three structural reasons.

The category is growing but not exploding. Mordor Intelligence sizes GCC quick commerce at $3.76bn in 2025, reaching $12.43bn by 2031 at a 22.05% CAGR.¹¹ Other houses are far more conservative — MarkNtel puts UAE quick commerce growth at 3.72% CAGR for 2026–2032.¹² That is a very wide disagreement, and it should be read as a warning: nobody actually knows, and a strategy that only works at 22% growth is a strategy with a single point of failure. A store-conversion strategy works at 3.72% too, because it did not require new capital to begin with.

The incumbent platforms are already profitable, which sets the bar. talabat reported Q1 2026 GMV of $2.685bn, of which the GCC was $2.122bn, with adjusted EBITDA of $130m at a 4.8% margin — while deploying around $25m in the quarter as part of a $120m 2026 investment programme spanning talabat mart and new retail.¹³ Delivery Hero's quick commerce GMV passed €7.5bn in 2025, growing over 30%, and is guided to approach €10bn in 2026.¹⁴ Any grocery retailer entering this competes with operators who already clear a positive margin.

Retail density is unusually high. Gulf cities are built around large-format grocery anchors with established catchments. That is precisely the asset base a store-based model monetises, and precisely what a new-build dark store has to compete against from a standing start.

Choosing between the two

Three questions settle most cases.

1. Do you already own retail assets in the catchment? Convert them first. Capex-light beats capex-heavy whenever the service level is the same, and in dense urban catchments it usually is.
2. Is there demand density but no asset? That is a genuine dark-store catchment. Build it — with a disciplined assortment and a tight delivery radius, treating it as infill in a wider network rather than as the network's default unit.
3. Is the site being proposed because a competitor opened one? That is not a business case. Competitive response is a reason to move quickly; it is not a reason to sign a ten-year lease across the road from a shop you already own. The 26%-to-44% overlap figure above is what that reasoning looks like at industry scale.

Speed is a product feature. Rent is a business model. The operators who win the next phase of quick commerce will be the ones who never confuse the two.

What is not public

A page that only presents the convenient numbers is not worth reading. Several things this argument would benefit from are simply not disclosed by anyone, and it is worth naming them.

• Nobody publishes cost per order. Not Blinkit, not Instamart, not Zepto, not Ocado. What exists is margin as a percentage of order value, on differing bases. Any specific cost-per-order figure circulating publicly is either modelled or invented.
• There is no credible published comparison of store-picking versus dark-store cost per order. The closest academic work is Calzavara, Finco, Persona and Zennaro's cost model in the International Journal of Production Economics (2023), which explicitly compares in-store picking, dark stores and central distribution — but its figures sit behind a paywall and are modelled rather than observed.¹⁵ Everything else in circulation is vendor marketing.
• Ocado does not disclose an automation payback period or an absolute cost per order, only cost as a percentage of sales.
• Store-based operators disclose least of all. Tesco publishes Whoosh growth and reach but no profitability. This is the awkward part of the argument: the model I am defending is the one with the thinnest public evidence, because it sits inside retailers who report it as part of a larger business rather than as a standalone venture.

Which is why the Kroger write-off matters so much. It is the one place where a company was forced to put a number on the difference — $2.6bn — and then report what happened next.

Common questions

Can a supermarket genuinely deliver in twenty minutes?

Yes — with a disciplined online assortment, designed pick paths, and rider dispatch modelled against the store's own peak. The binding constraint is almost never the driving. It is the minutes between order receipt and the bag reaching the dispatch point.

Is quick commerce profitable yet?

Barely, and only at extreme scale. Blinkit cleared 0.6% of net order value in the quarter to June 2026 with 2,443 dark stores. Instamart was at −1.8% of gross order value in the quarter to March 2026, improving to −1.1% on the March exit rate. Zepto lost INR 5,905 crore in the year to March 2026. The category has proven the demand and has not yet proven the model.

Does store picking annoy in-store customers?

It can, and that is a design failure rather than an inevitability. Congestion is a function of when picking happens and how paths are routed. Handled well it is invisible to shoppers; handled badly it is the first thing they mention.

Is this only viable for large-format stores?

No. Format matters less than catchment density and assortment discipline. A well-run compact store in a dense catchment frequently outperforms a hypermarket serving a thin one.

Are dark stores dead?

No. They are a precision tool that was misused as a default. Where there is real demand and no retail asset, a dark store is the correct answer. The error was treating it as the standard network unit rather than as infill — and the Indian overlap data suggests that error is still being made.

What is the single biggest mistake in conversions?

Offering the whole range online. It feels generous, it demos well, and it quietly destroys the pick economics of every order in the system.

---

---

Dipankar Biswas leads quick commerce and online supermarket operations for a major grocery retailer in the UAE, running a multi-country, multi-channel fulfilment network across own rapid delivery, scheduled online grocery and marketplace partnerships — 65+ stores across three fulfilment channels. He previously designed and launched 20 integrated dark stores across two countries, sited within existing hypermarkets, and led operations at noon and Swiggy. He writes on store-based quick commerce at dipankarbiswas.me.

Related: Most dark stores should never have been built — the argument, in short form.