Ghost kitchen: the mistakes that turn it into a commission business, and the honest alternatives

When a pure ghost kitchen stops being enough?

A pure ghost kitchen stops working the moment its fixed monthly cost, somewhere between 1,800 and 3,500 dollars for a module of 12 to 16 square meters, hangs on a single discovery channel you do not own.

The number that gives it away sits in the aggregator's weekly report, not in your P&L: when the share of orders arriving through organic search inside the app drops two weeks running and you patch the gap by buying position, you no longer run a restaurant, you rent traffic. Online food delivery reached 1.51 trillion dollars in 2026 and grows at a 6.24% compound annual rate through 2031 (Statista 2026), so the channel expands; your slice of it does not have to. That neighbor near the steakhouse paid 3,200 dollars a month for fourteen meters and closed in month seven, on the same average ticket as the operator who survived.

A virtual brand inside your existing kitchen: who it fits, what switching costs

Launch a virtual brand inside your own kitchen if you already have revenue, rent covered by the parent restaurant, and time slots where equipment sits idle. That is the profile: an operator with an active location, with a fryer, griddle or oven doing nothing between 3:00 and 6:00 p.m., and a menu that can split without buying a single new machine. Switching cost is mostly desk work — product photography, aggregator onboarding, listings, dedicated packaging — and it rarely exceeds 1,200 to 2,000 dollars of upfront investment, with zero construction. A steakhouse in northern Bogotá opened a wings brand exactly this way, and nine months later it bills 41 million pesos a month with real contribution margin, because the rent was already paid. If the algorithm reshuffles results, that brand loses sales; it does not lose the business.

Hourly shared kitchen versus a dedicated module

Renting kitchen time by the hour instead of signing for a dedicated module turns a fixed cost into a semivariable one, and that single difference decides who survives a bad quarter. It suits the operator without premises who wants to test a concept before committing capital: you buy blocks at 20 to 40 dollars an hour, cook three or four shifts a week, and walk away with no penalty when the concept misses. A dedicated module at 1,800 to 3,500 dollars a month demands that you cover the figure in the month the aggregator dims your visibility and equally in the month things go well. With the global dark kitchen market projected at 171.3 billion dollars by 2033 (Global Growth Insights), signing long looks tempting. I would sign short: flexibility beats the loyalty discount when your discovery is borrowed.

Pickup and direct ordering: taking the channel back

Direct ordering with in-store pickup is the alternative that hands the customer back to you, and the reason is arithmetic: you stop paying 18% to 30% commission per transaction and you keep the phone number, the email and the purchase history. According to Nick Kokonas, co-founder of Alinea Group and founder of Tock, relying on third-party platforms for discovery is the most expensive way to acquire a customer, because the operator pays for access to an audience that was already theirs; commission does not buy a customer, it rents an order. The profile that wins here is any business with weekly recurrence — office lunches, weekend family meals, standing plans — where the second order needs no discovery at all. Switching effort is real: your own catalog, a payment gateway, and months of insistence to shift barely 15% or 20% of volume off the app.

The asset is not the oven, it is the Google listing

For a kitchen with no storefront, the asset is its local digital engine, not the cooking equipment. With no sign, no window and nobody walking past, the only route to discovery is an in-app listing or a Maps search, and there a Google Business Profile with no visitable address and no photos of the place competes at a disadvantage against a venue holding 400 accumulated reviews. A restaurant with a street front runs three engines — foot traffic, Maps and aggregators; a ghost kitchen keeps one and a half. Diego F. Parra keeps pressing the point at Masterestaurant: measure that asset with the same discipline you apply to food cost. If 76% of U.S. operators believe technology gives them a competitive edge (National Restaurant Association 2024), the question is not how much technology you use, but which part of it belongs to you.

Your own delivery with a per-trip outsourced fleet

Paying for delivery by the trip, without surrendering your catalog, is the middle path almost nobody executes properly. You keep your digital store, collect the payment yourself, and pay 3 to 6 dollars per drop to an external fleet instead of a percentage that climbs precisely when your average ticket rises. The math reorders everything: on a 40-dollar ticket, 25% commission means 10 dollars, while the trip costs half that. It serves operators with volume concentrated inside a short radius and a high ticket; it punishes anyone selling 12-dollar orders scattered across the city, where fixed cost per drop eats the margin whole. Automation pushes on the same spot: Serve Robotics announced a rollout of up to 2,000 robots with Uber Eats (Form 8-K, SEC 2024), and that cheaper last mile reaches first whoever already controls the catalog.

The paradox of volume that never becomes cash

Every ghost kitchen operator meets the same tension late: the aggregator delivers the volume you need to justify the rent, and that very volume blocks you from building the owned demand that would set you free. It resolves by sequence, not by choice. First treat the platform as paid acquisition, with a declared ceiling on commission over total sales — I would cap it at 35% — and in parallel push every order toward your direct channel with a physical insert, a repurchase code and saved contact details. Uber Eats moved roughly 74.6 billion dollars in gross bookings during 2024 (Uber Technologies, Form 8-K, SEC); nobody competes with that reach, and nobody has to. What you need is that three years from now, dropping two ranking positions costs you a slow week rather than the lease.

When NOT to change anything?

Sometimes staying put is the right call, and it deserves saying plainly.

If your ghost kitchen already clears break-even with room to spare, if 60% or more of orders come from repeat customers who search your brand name inside the app, and if occupancy cost stays under 10% of sales, you do not have a structural problem: you have a business that works on somebody else's channel and you know it. Changing there costs money and attention. What I would do, without touching the model, is armor the flank: claim and complete the Google listing, stand up your own catalog even if it sells little in the first months, and store contact details from every order starting tomorrow. The day the algorithm shifts, that email list will be the only thing that did not change owners.

Where these routes genuinely split?

What separates a pure ghost kitchen from a virtual brand riding an existing kitchen is not operational, it sits on the balance sheet:

the first creates a brand-new fixed cost of 1,800 to 3,500 USD a month that only aggregator volume can pay, while the second consumes capacity that is already funded. When Rappi or Uber Eats reshuffles results (and that happens several times a year) the virtual brand loses sales; the leased module loses the business. Discovery behaves differently in each model. A restaurant with a visible storefront stacks three engines: foot traffic, Google Maps and aggregators. A ghost kitchen keeps one and a half, since a Google Business Profile with no visitable address and no venue photos competes badly against neighbors holding 400 reviews. Geotargeted advertising, an accelerator for a street-level restaurant, becomes oxygen here: switch it off and revenue drops the same week. Delivery unit economics punish each model differently depending on who pays the rent.

Where these routes genuinely split — in practice?

At a 14 USD ticket, 25% commission and 30% food cost, contribution margin lands near 4.90 USD per order. A virtual brand needs roughly 200 monthly orders to cover packaging, marginal labor and ads;

a leased module needs more than 1,100 just to reach break-even. Identical unit of business, two completely different floors. There is a genuine tension in this trade: the aggregator is at once your most expensive acquisition channel and your largest storefront. You resolve it by assigning roles, not by picking sides. Let the app acquire new customers and book that cost as marketing rather than cost of sales; let direct ordering, WhatsApp and the dining room retain those customers at full margin. Owners who file commission under cost of sales end up hating the platform. Owners who file it under acquisition know exactly how much a customer is worth, and when to stop paying.