Ghost kitchens: the numbers before and after the Masterestaurant method

Channel size never explains who makes money

A USD 1.4 trillion channel distributes sales, never margin, and that distinction decides a virtual kitchen's year. Statista counted roughly USD 353 billion in online delivery in the United States and around USD 450 billion in China during 2024, while Momentum Works measured USD 19.3 billion in Southeast Asia growing at 13%, figures that hand you the false comfort that any concept sells. It sells, of course. The point sits elsewhere: with a platform commission taking between 18% and 30% of the ticket before you pay the first hour of kitchen labor, a demand wave only amplifies whatever sign your margin already carried. If the order is born at minus three points, a thousand more orders a month will not fix it, they will multiply it by a thousand. Before you study the market, study the cost sheet of one single order.

Platform concentration: how much bargaining power is left?

Almost no operator negotiates commission, and the concentration data explains why. Earnest Analytics put Uber Eats at 26.1% of the US delivery market at the close of 2024;

Statista credited iFood with 87% of e-food bookings in Brazil in 2024; Ken Research measured Glovo near 31% and Just Eat around 26% in Spain for 2025. A market where one player touches 87% is not a market, it is a toll booth, and the operator who plans a virtual kitchen assuming preferential terms will arrive with good volume is planning on an illusion. The operating consequence is blunt and simple: your digital menu price has to be born already loaded with the highest commission in your city, not with the one you hope to win in next quarter's meeting.

Kitchen density is competition, not validation

Once a thousand ghost kitchens open in your city, the price stops being yours. CANIRAC counted more than 1,200 active dark kitchens in Mexico City in 2025, 40% above 2023, and Grand View Research calculated that the independent segment held 61.7% of cloud kitchen market revenue in 2025. That 61.7% usually reads as good news for the small operator, and I got it wrong for years: it also means your competition is not a slow chain but thousands of nimble kitchens that copy a menu in two weeks. Diego F. Parra insists at Masterestaurant on reviewing the density of your own postal code before any national market study, because 40% growth in two years turns into price pressure inside a three-kilometer radius, which is the only radius your rider covers in under twenty minutes.

Packaging and transit loss: the points nobody costs

Between 4 and 7 margin points per order disappear when packaging has no cost sheet. The carton, the lid, the bag and the seal add USD 0.55 to USD 1.10 per delivery, and that number rarely enters the costing because it arrives on a supplies invoice the chef never reviews. Run it against the case that opens this piece: 1,900 orders a month at an average USD 0.80 is USD 1,520, enough to pay a full head chef across much of Latin America. Add transit loss — the portion delivered cold, spilled or returned on a complaint — and the digital order stops resembling the dining room order. Concrete decision: open a cost line called packaging inside the food cost of every delivery dish, with its weight and its supplier, this week.

Multi-brand on one line: the arithmetic that holds margin

Three brands over one fixed cost pay it in thirds, and that is the only lever I have seen survive a commission hike. Rent, hood, walk-ins and a shift lead do not move whether you produce 60 or 160 orders a day, so every added brand sharing inventory and line dilutes the fixed cost without doubling payroll. Yet sequence matters more than the idea: one brand first, production validated, second brand in month four. What happens if you launch all three on day one? With three new menus, the shift lead masters none, dispatch times slide from fifteen to thirty minutes, the platform drops your ranking for lateness, volume falls, and you end up paying the same fixed cost on less revenue than with a single brand. Shared overhead rewards discipline and punishes impatience.

Automation and last mile: what is real and what is not yet

Delivery technology already moves millions of orders, though almost none of them are yours yet. Starship completed 5.8 million autonomous deliveries during 2024 according to Forbes; White Castle passed 100 drive-thrus with voice AI by late 2024, at a 90% order completion rate and roughly 60 seconds per order according to SoundHound as reported by Restaurant Dive; Grand View Research assigned North America 40.8% of the food robotics market. Read that as a direction signal, not a purchase plan: an operator running two or three brands does not amortize a robotic arm, it amortizes a unified ticket screen and a standardized recipe. The automation that actually pays inside a small virtual kitchen today is the data kind, because it tells you which dish loses money on each platform.

How to read these numbers in YOUR operation?

Benchmarks only work once translated to your scale, and there are three distinct scales. Small operation, under 900 orders a month and one brand:

your single priority is the packaging sheet and a channel price loaded with the 30% commission, because those USD 0.55 to 1.10 per order are your entire margin. Mid-size operation, between 900 and 2,500 orders and two brands — the USD 42,000 and 1,900 orders case — measure contribution margin per brand and per platform, never consolidated, or the losing brand hides behind the winning one. Group with three or more kitchens: negotiate supplies by volume and watch local density using the CANIRAC figure, that plus 40% in two years, before opening a fourth kitchen inside a saturated postal code. Pick your row and act only on that one this month.

Where these benchmarks come from and where they fail?

Every figure here comes from public sources with a year and an organization, and it helps to know what they miss. Market size data comes from Statista, Grand View Research, Momentum Works and Ken Research;

share data from Earnest Analytics 2024 and Statista Brazil 2024; local density from CANIRAC 2025 as reported by the Mexican press. Honest limits: share figures shift depending on whether they measure orders, bookings or billings, and one methodological difference moves several points; dark kitchen counts rely on trade registries that underestimate informality; none of it replaces your own margin per dish. Masterestaurant uses these benchmarks as a decision frame and never as a substitute for your own costing, the only number measured in your own till. Pull tomorrow your platform commission report for the last 90 days and compare it dish by dish.

What actually separates a losing ghost kitchen from a winning one?

The first gap is boring and accounting-shaped:

anyone running ghost kitchens without a packaging spec sheet donates four to seven margin points per order, because the box, lid, bag and seal add USD 0.55 to USD 1.10 that never entered the costing. On USD 42,000 of monthly sales, that gap alone pays a full kitchen manager. The second gap is structural. One brand on a kitchen's fixed cost pays 100% of the rent out of its own ticket; three brands on that same base split it in thirds. No marketing trick beats that arithmetic, which is why operators who really grasp the cloud kitchen model launch one brand, prove the production, then add the second around month four rather than on opening day.

What actually separates a losing ghost kitchen from a winning one — in practice?

The third gap is pricing, and it draws the most pushback. If your digital menu costs exactly what the printed one costs, you are funding the platform commission from your own margin.

The fix is not a blanket 20% increase; it is rebuilding the delivery menu around items that survive transit, with a price architecture of its own, which is the work Diego F. Parra runs at Masterestaurant before any ghost kitchen opens. The fourth gap is territory. Most ghost kitchen closures I review are not kitchen failures, they are zone failures: cheap rent in an area where average delivery time runs past 34 minutes, and the app punishes that in the ranking. So the correct order is territory first, brand second, menu third, never the reverse.