What is a dark kitchen: the mistakes that sink it and the method that holds it up

What exactly is a dark kitchen?

A dark kitchen is a production site with no dining room, no storefront and not a single customer inside, shipping exclusively through delivery aggregators and its own channel.

The deliverable of step one is a one-page sheet where you state three verifiable things: square meters devoted to production, active channels with their signed commission, and delivery radius in kilometers. If the word «tables» shows up on that sheet, you are not building a dark kitchen but a restaurant with the lights off in front, which is a different business. The format already carries weight: Statista puts ghost kitchens at roughly 15% of US foodservice delivery sales in 2023, and user penetration in meal delivery reached 27,5% in 2024. The asset is not the square footage: it is your POSITION in the listing. Before signing any contract, calculate contribution margin per order after commission, packaging and paid promotion; that figure, not rent, decides whether the model survives.

Run contribution margin per order before you sign

Take your average ticket, subtract food cost, subtract the aggregator commission —DoorDash publishes plans of 15%, 25% and 30% according to CloudKitchens— and subtract real packaging, not the supplier estimate. An operator in Medellín cut rent from 4.800 to 1.400 dollars a month and closed in month seven: his 9-dollar ticket could not carry 27% commission plus 2,3 dollars of packaging. The deliverable is a table with three ticket scenarios —low, medium, high— and the contribution in dollars for each one. If the medium-ticket cell does not clear 2,5 dollars, the business does not exist; what exists is the illusion of cheap rent. A ghost kitchen location is chosen by how many active app users live inside the delivery polygon, and only then by the price of the lease. Draw a three-kilometer radius on the map, open Rappi, Uber Eats and DiDi Food from an address inside that polygon, and count how many competitors in your category appear on the first two screens.

Pick location by demand density, not by rent per meter

Mexico shows the channel's scale: DiDi Food reported more than 360 million orders delivered over five years through 2024, and in Colombia meal delivery penetration stood at 19,8% in 2024 according to Statista Market Insights. What this step leaves behind is a map with the polygon drawn, the competitor count by category and the courier's estimated travel time. A lease 40% cheaper ten kilometers away from the customer is the most expensive mistake in this sector. A dark kitchen menu gets designed backwards from prep time, because the algorithm punishes every minute above the local average. Choose between eight and twelve SKUs sharing at least 60% of their inputs, time with a stopwatch how long it takes from order in to sealed bag out, and drop any dish that runs past 14 minutes at peak. The Medellín case sank at 21 minutes of preparation: that pushed him outside the range where Rappi shows you on the first screen.

Design the short menu that fits in 14 minutes of prep

The deliverable is a card per dish with food cost, timed preparation and how the packaging behaves after 25 minutes on the road. We work with a ceiling of 32% food cost per dish, and in delivery that ceiling drops, because the commission lands on top of it. A virtual brand lives in the aggregator listing, and that listing is built the way a display case is built: real photography of every dish, descriptions with ingredients, price adjusted to the channel. Shoot your ten main dishes on a neutral background, write descriptions of fifteen to thirty words, and raise the channel price 15% to 20% above your own price to absorb the commission without wrecking the margin. Given the scale platforms move —DoorDash closed the fourth quarter of 2024 with 685 million orders, up 19% year over year— that listing is your only counter. The deliverable is a published, approved listing on each platform, with real operating hours and a properly chosen category.

Build the virtual brand and its listing on every aggregator

Verification is simple: order from your own menu using a different account and look at what arrives. The mistakes that kill a ghost kitchen repeat with an almost boring monotony, and all of them surface between month five and month eight. First: treating the dark kitchen as a cheaper restaurant when it is a DIFFERENT business, with another funnel and another law of growth. Second: paying for aggregator promotion without measuring contribution after the discount, which usually leaves the order underwater. Third: accepting more than two virtual brands in the same kitchen before the flow is stable, because prep time explodes. Fourth: never touching the packaging, when packaging explains 2,3 dollars of the variable cost. At Masterestaurant, with twenty years inside kitchens across 43 countries, Diego F. Parra insists on an order almost nobody respects: contribution first, volume after. Reversed, it burns the entire working capital. Your place in the listing is decided by three variables you control: preparation time, customer rating and cancellation rate.

The three metrics that govern your place in the listing

Measure daily prep time against the category average in your zone, hold the rating above 4,6 and keep cancellations under 2% of accepted orders. None of the three gets fixed with an advertising budget, and there sits the paradox of the channel: you pay an intermediary between 18% and 30% of every sale and that intermediary also decides who sees you, so the only real lever is operational. The deliverable is a weekly board with those three figures and the variation against the previous week. Statista projects meal delivery penetration at 29,2% for 2026: more demand is coming, and more competitors stacked above you as well. You know the dark kitchen was built right when you can answer six questions with a number in hand, not with an impression. One: contribution margin on the medium ticket, after commission, packaging and promotion, clears 2,5 dollars. Two: timed preparation at peak stays under 14 minutes.

Closing checklist: how to know everything is right

Three: your rating on every platform passes 4,6. Four: cancellation rate never touches 2%. Five: food cost per dish sits below 32%. Six: you ordered your own menu from another account and the packaging arrived intact after 25 minutes. If any cell comes back empty, do not open the second virtual brand and do not sign the lease extension. Fix that cell this week and measure again the following Monday, with the sheet in front of you. The first difference is one of nature: a restaurant buys LOCATION and a dark kitchen buys VISIBILITY. You are not paying for the corner lot, you are paying to appear among the first eight results when somebody 2.4 kilometers away opens the app hungry, and that position is earned through short prep times, high ratings and low cancellation rates, three variables the owner controls and almost none of them measure. Second comes cost structure.

The four differences that decide the cash

A restaurant with a dining room charges rent, service payroll and utilities against break-even; the dark kitchen charges aggregator commission against EVERY order, which converts a fixed cost into a variable one and rewrites the math of volume. Doubling sales in a restaurant improves margin; doubling delivery sales with a mis-costed commission simply multiplies the loss. Third is customer ownership. In a dining room the customer is yours: you greet them, recognize them, ask for their email. Inside delivery aggregators the customer belongs to the platform and you rent access. According to Tomer Molovinsky, senior restaurant analyst at J.P. Morgan, reliance on third-party channels erodes operators' pricing power because the intermediary owns the relationship and the repeat-purchase data. That is why the method demands moving a third of sales into an owned channel before month twelve. Fourth is correction speed. A restaurant spots a menu problem after four weeks of comments; a virtual brand spots it within 48 hours by reading per-dish ratings and repeat rates.

The four differences that decide the cash — in practice

That speed is the format's genuine advantage, and it gets wasted when the owner opens the platform dashboard only to check how much was deposited.