How to start a dark kitchen from scratch: the expensive mistakes against the method that leaves margin

Why does a dining-room menu price wreck a dark kitchen before it opens?

Because a dine-in price cannot absorb marketplace commission, and that commission runs 15 % to 30 % with an effective cost reaching 30 %-40 % per order (Food On Demand 2026).

Run the math on a 6 USD dish listed in-app: at 27 % commission —the band independents pay on Uber Eats, 27 % to 30 % according to eLogii 2024— you keep 4.38 USD net. If the input costs 2.10 USD, your real food cost is not the 35 % you wrote on the sheet, it is 48 %. The expensive path inherits the dining-room price and finds the commission at month-end close; the correct method puts commission into the recipe card before the menu exists. The second one wins, and not narrowly: it is the difference between billing 30,000 USD a month with cash in the bank and billing the same with nothing to cover payroll. Forty to sixty square meters win, because in a kitchen without diners the spare meter produces zero orders and still gets paid twelve months straight in rent, utilities and extraction.

Forty-five square meters against one hundred twenty: the rent that dispatches nothing

A 120 m² unit with two hot lines dispatches roughly the same per peak hour as a well-laid-out 45 m² one —the bottleneck is the griddle and the fryer, never the aisle— yet it triples fixed monthly cost. With industrial-zone rents between 8 and 14 USD per meter across the region's main capitals, those extra 75 meters mean 600 to 1,050 USD monthly you must cover with 9 USD tickets carrying 2 USD of net margin: three to five hundred additional orders every month just to break even. The expensive path buys meters thinking about growth. The correct one buys meters thinking about the month-four break-even point. Two virtual brands on the SAME mise en place beat both the single brand and the five-brand catalog, and the reason sits in inventory, not marketing. With one brand you ride a single demand curve; with two sharing 70 %-80 % of inputs —chicken, potato, bread, mother sauces— you double marketplace exposure without adding one purchase reference or one meter of cold storage.

One brand or two: the mise en place decides, not the logo

The repeated mistake is launching five brands with different menus: fifteen new references, slow rotation, waste eating the very margin those brands were supposed to bring. Diego F. Parra orders it this way in the kitchens Masterestaurant works with: recipe card with the net price first, equipment next, brand design only at the end. Reversed, you pay for a pretty logo sitting on a dish that loses money on every delivery. Your own channel wins on margin and loses on volume, which is why the right answer is to blend rather than choose. DoorDash pickup commission sits at 6 % (CloudKitchens 2024) against the 27 %-30 % independents pay for Uber Eats delivery (eLogii 2024): twenty-one points of spread on the same dish. But that 6 % rides on a traffic base you have to build yourself, while Uber Eats moves close to 95 million users (Uber Technologies 2024) and DoorDash grew marketplace GMV 20 % year over year in 2024.

Your own channel at 6 % against the app at 27 %: where the hidden margin lives

The operation I recommend: open on marketplaces to buy expensive demand for the first ninety days, and from order one slip into every package the coupon that migrates the customer to your own channel. If by month eight 25 % of your orders arrive with no third-party commission, you recovered the equivalent of five food cost points. A 52 m² kitchen in an industrial zone opened on 47,000 USD of total investment and closed month seven with positive cash, and the breakdown of where the money went explains the whole method. The split: 18,000 in equipment (griddle, two fryers, a used blast chiller, cold room), 9,000 in fit-out and hood, 6,500 in deposit and first rents, 7,000 in working capital and 6,500 in photography and the build-out of two brands. Average ticket landed at 9.40 USD with 28 % commission, meaning 6.77 USD net; with 2.35 USD food cost and labor at 29 % of revenue —inside the 25 %-35 % range the Bureau of Labor Statistics reports— every order left 1.90 USD of contribution.

The case: 52 m², two brands and positive cash by month seven

Break-even fell at 1,640 monthly orders. He hit it in week twenty-seven, not on a spike, but because he never added a third brand. If your recipe card was built with commission inside, three extra points cost you 0.28 USD on a 9.40 ticket and you adjust the in-app price by thirty cents without touching the recipe. If it was built on the dining-room price, those three points push contribution margin from 1.90 USD down to 1.62, break-even jumps from 1,640 to 1,920 monthly orders, and come April you discover you need 17 % more volume just to stand where you stood in January. That is exactly where the temptation to shave portion weight appears, and shaving weight is how you lose your app rating and with it the ranking that took three months to build. The conclusion comes before the argument: commission is not an expense, it is an input, and you cost it like cooking oil.

What happens if the marketplace raises your commission three points next year?

A kitchen that gets this survives the annual platform renegotiation without flinching. Automation stopped being a corporate luxury and now pays for itself in a 50 m² kitchen, which is precisely what was not true four years ago.

A dashboard that consolidates Rappi, iFood and your own channel into one view, forecasts demand by time slot and fires an alert when a dish's contribution margin drops below threshold, runs 80 to 200 USD monthly against the 450-600 USD of a part-time cook. And the market already moved: 86 % of operators say they are comfortable using AI (Toast 2025). The limit has to be stated plainly, because foodtech gets sold as if it fixed everything: a dashboard shows you the dish loses money, it does not repair it. If the recipe card is wrong, AI will report with two-decimal precision how much you bleed every Sunday.

What to choose by profile: three scenarios, three decisions?

Under 60,000 USD and aiming for positive cash before month eight, the correct method admits no argument:

one kitchen of 40 to 60 m², two brands on the same mise en place, commission inside the recipe card and your own channel seeded from order one. If your capital sits between 60,000 and 120,000 USD and you already run a physical restaurant, use the existing kitchen during off-peak hours before signing a second lease: it saves 8,000 to 12,000 USD of fit-out and buys three months of real demand data before you commit to meters. If you start under 30,000 USD, do not build your own kitchen yet; enter a dark kitchen hub at 1,200-1,800 USD monthly all-in and validate the menu with someone else's money. This week do one thing: recalculate your three best sellers with commission deducted.

What to choose by profile: three scenarios, three decisions — in practice?

If any crosses 38 % net food cost, do not publish it. The founding difference is the order of decisions. On the expensive road the price is inherited from the dining room and the commission gets discovered afterwards;

in the method, commission enters the recipe card before a menu exists. A 6 USD dish on the app at 27 % commission leaves 4.38 USD net, and if the ingredient costs 2.10 USD the real food cost is not 35 % but 48 %. That single line explains why a dark kitchen can bill 30,000 USD a month and still miss payroll. Footprint is the second fracture. Every extra square metre is paid in rent, utilities and extraction across the full twelve-month lease, and in an operation with no guests it generates zero additional revenue. Forty-five well-planned metres ship more orders per hour than a badly designed hundred and twenty, because the bottleneck in a dark kitchen is never floor space: it is the packing station.

Where the math breaks and why?

Brand count separates the owner who understands the marketplace from the one who merely uses it. With one brand you appear in a single listing;

with three sharp ones —executive lunch, late-night comfort food, desserts— you appear in three, and the same cook produces all three on the same mise en place. You are not multiplying the kitchen, you are multiplying the shop window. Reading speed decides whether you fix things in time or perform an autopsy. Margin reviewed every 24 hours lets you pull a dish that lost profitability in week three; a monthly spreadsheet tells you in week seven, after you bought ingredients for that dish twice more. I got this wrong for years: I treated the dashboard as a chain-restaurant luxury, and it turned out to be the cheapest line of the whole build. Here is the tension few owners resolve well: the marketplace gives you immediate volume and takes your margin, while the direct channel gives you margin and no volume.

Where the math breaks and why — in practice?

The answer is not to choose.

Launch on Rappi and iFood to fill the kitchen from month one, then use that traffic to build your own base through packaging, codes and repeat offers, pushing direct sales from 5 % to 25 % of revenue by month twelve.