How to start a dark kitchen from scratch: the mistakes that burn cash and the method that closes the numbers

Step 1: set your in-app selling price before you touch a single recipe

The first number you write down is not the dish cost but the price the customer will see in the app, because every other figure comes out of that price and not the other way around, which is how a dining-room restaurant gets costed. If you settle on 12 USD for your anchor dish, that is the ceiling from which everything else gets subtracted, and at a 27% commission the platform keeps 3.24 USD before your kitchen even fires up a burner. The deliverable here is a sheet with three columns per dish, in-app price, applied commission and the real net that lands in your bank account, and you verify it against the aggregator's actual payout in week two. Write it with your direct competitor's menu open in another tab, because in delivery the comparison sits one scroll away.

Step 2: subtract commission and packaging before you calculate food cost

Commission and packaging come off the TOP, glued to revenue, and only the remainder feeds your food cost calculation, which must never exceed 32% of the selling price. On those 12 USD, with 3.24 USD of commission and packaging running 0.45 to 0.80 USD depending on whether you ship soup, sauce or something fried, you are left with roughly 7.96 USD, of which raw material can take 3.84 USD at most. Contribution margin per order lands at just over 4 USD, and that number, not the rent, governs the entire model. Booking commission down below, next to stationery and the accountant's fee, is the classification error I have corrected most often in foodtech financial models, because commission behaves exactly like food cost, it climbs when you sell more.

Step 3: translate your fixed structure into orders, not square meters

A dark kitchen's break-even gets measured in orders, never in covers or table turns, and that translation separates a model that holds up from a napkin full of good intentions. Add module rent, utilities, kitchen payroll, management software and admin, and assume 3,400 USD monthly of fixed structure for a small two-person-per-shift operation. At a contribution margin of 4.10 USD per order you need 830 orders a month just to break even, roughly 28 orders a day without a single day off. The deliverable is one figure written on page one of the model, break-even orders, verified every Monday against actual orders accumulated so far that month. If the number looks big to you, that means you are finally seeing the business as it is.

Step 4: build the short menu for the channel, not for the dining room

A profitable dark kitchen menu lives between 12 and 18 references sharing ingredients, because every extra dish loads inventory, waste and dispatch time onto a kitchen that gets charged per transaction. Diego F. Parra insists at Masterestaurant on an uncomfortable criterion, that the dish photographing best is not always the one that should lead the menu, and in delivery the winner is whatever arrives intact at minute 25 inside a sealed container. Fries lose texture, breaded items sweat, emulsified sauces break. Build a simple matrix with four columns, contribution margin in currency, weekly turnover, travel resistance and packaging cost, and cut without sentiment anything scoring low on two of the four. The menu is closed when every reference shares at least one ingredient with two others.

Step 5: build the physical operation with the minimum viable layout

A ghost kitchen runs well between 20 and 40 square meters when the flow moves in a straight line, goods receiving, refrigeration, mise en place, hot line, packing and courier handoff, with no two people crossing paths. This channel rests on a market that is now structural rather than a passing fashion, since Circana puts roughly 75% of restaurant traffic off-premises, and the National Restaurant Association reported in 2025 that 65% of limited-service operators offer delivery. What you must not skimp on is the packing station, a dedicated table with bags, seals, labels and a thermometer. The deliverable is a scale drawing with the six stations marked and a stopwatch reading of one complete order, which in a healthy operation should stay under 12 minutes of kitchen time.

Step 6: the four mistakes that blow up the model by month three

The most expensive mistake is not bad cooking, it is accepting co-funded promotions without recalculating margin, because a two-for-one stacked on top of a 27% commission turns a profitable order into a donation with logistics included. The second is failing to separate the bank account from aggregator payouts, which arrive biweekly and lagged, so the owner confuses cash flow with profit. The third is launching three virtual brands in the same month out of one kitchen, multiplying waste before the first one has stabilized. The fourth is pricing identically to the dining room, ignoring that no platform was taking almost a third there. If your orders grow 40% and profit does not move, the problem is not volume, it is that contribution margin per order was miscalculated from day one.

Step 7: close with the checklist that tells you everything landed

Your dark kitchen is ready to operate when you can answer seven concrete points with figures rather than impressions. In-app price fixed per dish and benchmarked against direct competitors. Real commission confirmed on the payout statement, not the one promised in the contract. Packaging costed per reference between 0.45 and 0.80 USD. Food cost measured with a physical inventory count and sitting below 32%. Contribution margin per order calculated in money, not in percentage. Break-even orders written and visible on page one of the model. Kitchen time clocked under 12 minutes at peak. If any of the seven is blank or estimated, you have not finished building the kitchen, you merely rented it. Tomorrow take the next eight orders, measure them one by one against this list and compare the result with what you wrote in the model.

The four differences that decide the outcome

Break-even moves from square meters to orders. In a dining room you think covers per shift and table turns; in a dark kitchen the only denominator that matters is the order, and with 4.10 USD of contribution margin per order against a fixed structure of 3,400 USD a month, you need 830 orders to avoid a loss. That figure, not the rent, belongs on page one of your model before you sign anything. Commission is not an administrative expense, it is a variable cost of sale. Booking it down below alongside stationery and the accountant is the classification error I correct most often in foodtech financial models. It belongs at the top, glued to revenue, because it behaves exactly like food cost: it rises when you sell more.

The four differences that decide the outcome — in practice

Put it in the wrong place and it paints a 68% gross margin that does not exist. Delivery packaging is an input, not marketing. Negotiated well it runs 0.45 to 0.80 USD, and operations push it to 1.10 USD chasing a pretty embossed box. That 0.40 USD gap across 900 orders is 360 USD a month, nearly the rent of a shared-kitchen module. Cap it hard at 5% of the ticket and negotiate quarterly volume. A virtual brand multiplies sales, not kitchen capacity. A second brand riding the same cooking line can add 25-30% more orders at zero additional fixed cost, and that is the real leverage in this model. But when the second brand demands another base input, another supplier and another cook time, you did not diversify: you split your kitchen in two and doubled the waste.