Ghost kitchen: the mistakes that sink one and the method that holds it
A ghost kitchen is a restaurant with no dining room that cooks only for delivery, and in 2026 it pays off only when the contribution margin per order covers the aggregator commission before anything else. With average US tickets of USD 20-35 (Lightspeed 2025) and commissions reaching 30% where no legal cap exists, costing a dish as if it were a dine-in plate shows up as a loss by month four. The right method starts with unit economics, never with the kitchen build-out.
An owner wrote to me in March with the arithmetic backwards: a twelve-month lease on a cloud kitchen station, three virtual brands loaded onto two aggregators, and a food cost calculated on the dine-in menu price rather than on the price customers actually see in the app after the platform markup. Revenue looked healthy. Every single order lost money, and it took four months to surface, because the platform paid weekly and deposited net of commission, so the figure he watched arrived already discounted and looked perfectly normal to him.
That pattern repeats, and it does not belong to careless operators. Global delivery moved from USD 380.43 billion in 2024 toward a projected USD 618.36 billion by 2030, a 9.0% compound rate per Grand View Research, while nearly 75% of US restaurant traffic now happens off-premise according to the National Restaurant Association. Demand is not the constraint. What rarely exists is the arithmetic that turns demand into cash, because this model shifts the cost of experience — room, servers, atmosphere — into a place that hides better: commission, packaging, and whatever the trip destroys.
At Masterestaurant we run the same framework across brick-and-mortar and virtual brands, and the practical difference is where break-even breaks. In a dining room, rent rules. In a ghost kitchen, the platform's percentage rules, and you do not negotiate that percentage — your volume does, or a regulator sets it. New York capped it permanently at 15% per delivery plus 5% for other services; San Francisco held it at 15% back in 2020. Across most of Latin America, no cap exists at all.
Ghost kitchen, side by side
| Typical mistake (improvised ghost kitchen) | Masterestaurant method | |
|---|---|---|
| Dish costing | ✕Food cost figured on dine-in price: a stated 32% becomes 41% real after a 27% commission | ✓Food cost ≤32% on the NET price that lands in the bank, commission stripped out first |
| Aggregator dependence | ✕95-100% of orders on one app; 25-30% commission, no cap, no customer data | ✓Target of 30% direct-channel orders by month 6; blended effective commission ≤20% |
| Number of virtual brands | ✕3-5 brands at launch; 42 SKUs crammed into a station built for 12 | ✓One brand until 400 sustained monthly orders; the second enters at ≥85% fill rate |
| Packaging and trip loss | ✕Packaging chosen on unit price: 6-9% complaint rate for temperature or spillage | ✓Packaging tested across a 25-minute trip; complaints ≤2%, packaging cost ≤4% of ticket |
| Location | ✕Cheap warehouse signed without measuring demand density or the 20-minute radius | ✓Territory prefeasibility: demand, competition and drive times settled before signing |
| Cash control | ✕Owner watches the weekly net deposit; per-order margin never gets calculated | ✓Contribution margin per order tracked daily; alarm triggers below USD 6 net |
| Break-even | ✕Estimated by feel; rent and payroll loaded onto the dish and distorting price | ✓Break-even stated in orders per day, payroll and rent outside the dish, inside the equilibrium |
Step 1: calculate contribution margin per order using the app price, not the dining-room price
Before signing anything, take one dish and subtract ingredient cost, packaging and the aggregator commission from the price published in the app: what remains is your REAL CONTRIBUTION MARGIN, and if that figure does not clear 45% no ghost kitchen survives. The owner I described above costed against his printed menu while charging through the app, and the platform ate that gap for four months without it ever showing up in a report. New York capped commissions permanently at 15% for delivery plus 5% for other services, and San Francisco limited them to 15% back in 2020; across most of Latin America there is no cap at all, so 25-30% is the prudent assumption. Deliverable: one sheet listing app price, food cost, packaging and commission for every dish on the virtual menu. Verify it by matching that sheet against the platform's weekly net deposit: if they do not reconcile to the cent, the sheet is wrong.
Step 2: set the target average ticket before designing the menu
Ticket size governs the menu and not the other way around, because commission is charged as a percentage while the fixed per-order costs —container, lid, bag, label— barely move between a USD 14 order and a USD 32 one. With tickets running USD 20-35 in the United States per Lightspeed 2025, a USD 14 order carrying 28% commission leaves under USD 4 after packaging, and that will not pay the shift of whoever assembled it. Define the target ticket first, lift it with two-person combos and a beverage, then write the menu. Markets like Colombia, at USD 1,180 million in 2024 with a 7.32% compound rate through 2029 according to Statista Market Insights, reward that discipline because volume grows while commission does not fall. Deliverable: a written target ticket with the combo mix that sustains it. Verify by measuring the real ticket over two weeks inside the aggregator dashboard.
Step 3: run one kitchen with virtual brands that share inventory
Three virtual brands only make sense when they share 70% of raw materials, because every extra SKU entering the walk-in becomes waste nobody notices until month-end inventory. A wings brand, a burger brand and a bowls brand can live on the same chicken, the same bread and the same base sauce; drop a sushi concept into that mix and you have added fresh fish with a two-day shelf life plus a knife nobody else touches. The global ghost kitchen market reached USD 70,400 million in 2024 according to Research and Markets, and much of that growth came from operators who multiplied brands over a single production line rather than from those who opened new kitchens. Deliverable: a shared-input matrix by brand with the overlap percentage calculated. Verify it when monthly waste drops below 4% of purchases and the count closes with no manual adjustments.
Step 4: negotiate around the commission by using the aggregator for acquisition
The platform taking your margin is the same one bringing the volume, and resolving that tension is not about fighting it but about deciding which order you give away and which one you keep. Let it bring the first; do not let it keep the third and the tenth. Slip a card with your own channel and an exclusive discount into every bag, because on a direct channel you pay no 25-30% commission and can hand half that saving to the customer without touching margin. Nearly 75% of United States restaurant traffic now happens off-premise according to the National Restaurant Association, so the volume exists; what gets decided is who owns the relationship. Deliverable: an active direct channel with a short link and a printed card in each order. Verify it when direct orders pass 20% of total volume within ninety days.
Step 5: treat packaging and in-transit waste as their own cost line
Packaging is not an administrative expense, it is disguised food cost, and at Masterestaurant we give it its own line because a fry arriving soggy costs twice: the replacement and the review. Budget it between 4% and 7% of selling price and track refunds for badly delivered orders separately, since in poorly built kitchens those exceed 3% of monthly revenue. A vented container costs a few cents more than a sealed one and drops fried-item returns close to zero, so the math pays for itself inside the first week. I got this wrong for years, recommending savings on packaging to protect margin, until refund data showed that saving was the most expensive item in the operation. Deliverable: packaging cost per dish and weekly refund rate on the same dashboard. Verify it when refunds stay below 1.5% for a full month.
Step 6: the mistakes that sink a ghost kitchen with strong sales
Four things sink this model, and none of them resemble bad cooking. First comes aggressive in-app promotion: every dollar discounted leaves twice, once from margin and once from the base the commission is calculated on, so 20% off against 28% commission can push the order negative. Second comes signing long space contracts, like the twelve-month deal that cost the owner in our case four months of losses. Third comes reading the platform's net deposit as if it were revenue, when it already arrives discounted and hides the exact number that needs watching. Fourth comes loading payroll and rent onto the plate: those belong in break-even, not unit costing, and mixing them produces inflated prices the app punishes with lower conversion. Deliverable: these four traps reviewed once a month. Verify it with margin per order holding steady for eight weeks.
Step 7: scale only when unit economics hold without promotion
Open the second kitchen once the first has produced positive margin for eight consecutive weeks WITHOUT a single active promotion, not before, because scaling an order that loses money simply multiplies the loss by the number of stations. Think through what happens if you open three sites at -USD 0.80 per order and reach 400 daily orders: that is USD 960 lost per day, USD 28,800 a month, and rising revenue will tell you the opposite story until the bank corrects it. The global delivery market moves from USD 380,430 million in 2024 toward a projected USD 618,360 million by 2030 at a 9.0% compound rate according to Grand View Research, and that demand rescues nobody whose arithmetic is unresolved. Deliverable: eight documented weeks of positive margin with zero promotions. Verify it by laying the aggregator's weekly reports side by side.
Closing: how to know everything is right
Your ghost kitchen is correctly built when you can answer six numbers without opening a file: contribution margin per order, real average ticket, effective commission percentage, packaging cost per dish, refund rate and share of orders coming through your direct channel. Diego F. Parra runs that same six-line dashboard with single-site operators and with twenty-unit groups, because the model does not change shape as it grows, only scale. Run the full checklist: a costing sheet built on app prices and reconciled against the weekly deposit, target ticket met two weeks running, input overlap above 70%, direct channel above 20% of orders, refunds under 1.5%, and positive margin across eight promotion-free weeks. If a single one of those six fails, the failure is already named and the matching step tells you where to go back. Today, take your best-selling dish and calculate its margin using the price the customer sees in the app.
Where the outcome actually gets decided?
The difference does not sit in the kitchen: it sits in who controls final price. In a physical restaurant you set the price and the guest pays all of it;
in a ghost kitchen the aggregator charges its percentage on the price you published, so every promotional dollar leaves twice — once from margin, once from the base the commission is computed on. Aggressive in-app promotion is the fastest way to break a business whose revenue line looks fine. One paradox deserves a straight answer: the platform eating your margin is also the platform delivering the volume without which the business would not exist. The resolution is neither fighting the aggregator nor abandoning it, but using it as acquisition rather than as the only channel. Let it bring the first order; do not let it keep the third and the tenth. The mix that works in the operations we review runs 60% to 70% aggregator with the balance direct, and that mix pulls blended commission from 27% down to roughly 19%.
Where the outcome actually gets decided — in practice?
Scale misleads too. With more than 20,000 operating ghost kitchen locations in the United States per Statista, the easy reading says the model is proven and entering is enough.
The correct reading runs the other way: at that supply level the edge is no longer existing, it is holding the lowest cost per order inside your radius, because an app customer compares twenty options on one screen and owes zero loyalty to a cloud kitchen they have never seen. And here is something that contradicts what gets repeated at foodtech conferences: if your concept leans on dining-room experience, menu narrative or a server's suggestive selling, a ghost kitchen is not a natural extension but a different business wearing the same logo. They coexist; they do not copy each other. When the operation keeps a physical location, the printed menu still governs the guest experience — service pace, narrative, upselling — while the QR menu and the delivery listing do another job entirely: price updates, accessibility, analytics. Both, each in its own role.
Head to head: improvisation against method
What the ghost kitchen that closes by month eight does
- Launches three or four virtual brands because loading them costs nothing, then chokes one station that misses ticket times at peak
- Copies dine-in prices into the app and hands the aggregator 27% of a margin that was already tight
- Signs the station lease before counting how many orders actually live inside the real delivery radius
- Judges the business by gross app revenue instead of net deposits or per-order margin
- Picks packaging on unit price, then finds at month three that 8% of orders generate a complaint
- Owns no direct channel, so when the platform raises commission two points there is nowhere to move volume
What the one still open in year three does
- One virtual brand until it sustains 400 monthly orders with fill rate above 85%
- Price built from net: commission out first, then food cost ≤32%, then everything else
- Territory prefeasibility before the lease: demand density, category competition, real drive minutes
- Daily dashboard with contribution margin per order and an alarm below USD 6 net
- Packaging validated on a real 25-minute run, thermometer in hand, using the dish that travels worst
- Direct channel pushed from day one, targeting 30% of volume by month six
The numbers that govern the model
“We opened with three virtual brands and 42 SKUs in a station designed for twelve, and at peak our ticket times stretched to 38 minutes. We cut back to a single brand and eleven dishes, and the average ticket climbed from USD 21 to USD 26 because we could finally build combos that travelled well. Blended commission fell from 27% to 19.4% once the direct channel reached 31% of volume in month six, and contribution margin per order went from USD 3.80 to USD 7.10. Same station, same equipment; what changed was the order in which we made decisions.”
The method, step by step, with its control number
Three things come before step one: six months of operating cash that does not depend on sales, your real commission confirmed in writing with each aggregator — the published rate and the rate applied to you rarely match — and a spec sheet per dish costed to the cent. Deliverable: a one-page document holding those three numbers. Checkpoint: if you cannot write your exact commission to two decimals, you do not have a business yet, you have an intention.
Take the price the customer sees in the app, subtract the aggregator commission, and calculate food cost on THAT net figure, where 32% is the ceiling rather than the target. Add packaging, which belongs at 4% of ticket or below. Whatever remains is your contribution margin per order. Deliverable: a sheet with margin per dish and weighted average margin. Checkpoint: contribution margin ≥USD 6 net per order, or the dish stays off the menu. Common errors here: loading payroll and rent onto the dish, when both belong to break-even rather than unit costing, and forgetting packaging entirely.
A cloud kitchen lives or dies inside its 20-25 minute radius, because beyond that food arrives cold and the complaint eats the margin. Measure household density, direct competition inside the radius, and actual peak-hour drive times rather than a Google Maps estimate at three in the afternoon. Our «territory intelligence» tool carries the prefeasibility load here, and the «gastronomic radar» tells you which categories are saturated in that zone before you commit to a concept. Deliverable: a zone report ranking three candidate sites. Checkpoint: ≥3,000 target households inside the radius and ≤4 direct competitors in your category.
The mistake I run into most often is multiplying brands because loading them into the app costs nothing. It costs ticket time, which is the single thing a delivery customer punishes without mercy. Open with one brand and a short menu built from dishes that survive a 25-minute trip and share ingredients, so inventory does not fan out on you. Deliverable: an 8 to 11 item menu with spec sheets and a documented trip test. Checkpoint: ticket time ≤18 minutes at peak and temperature or spillage complaints under 2%.
The aggregator brings the first order; your job is keeping the second. Put a concrete reason to order direct inside every package — not a generic discount, but something the platform cannot match — capture the customer record with permission, and measure repeat rate. Deliverable: an owned customer base with last-order date and frequency. Month-six checkpoint: ≥30% of volume through the direct channel and blended effective commission ≤20%. If you are still at 5% by month six, the customer is not the problem: you never gave anyone a reason to switch apps.
Payroll, rent and utilities never load onto the dish; they get covered by accumulated contribution margin, which makes orders-per-day-to-zero the number that matters. Calculate it, post it on the kitchen wall, and review it daily alongside per-order margin and direct-channel share. Deliverable: a three-number dashboard updated before every close. Checkpoint: break-even reached at ≤45 daily orders and theoretical-versus-actual food cost variance under 2 points.
And with AI?
Optimize channels, pricing and unit economics of your dark kitchen. Diego F. Parra is an expert in AI applied to restaurants.
Free tools: ghost kitchen
Ecosystem tools that hold the method together
These three pieces solve what a homemade spreadsheet stops handling once volume arrives: modelling the concept, projecting growth, and watching weekly cash, which in delivery is where any drift in per-order margin shows up first.
Diego F. Parra — International consultant, expert in creating and scaling restaurants and in AI applied to restaurants, foodtech and HORECA. Methodology applied in 8.400+ restaurants across 43 countries · Expert in Artificial Intelligence applied to restaurants, hospitality and food businesses · 20+ years in restaurants, catering, large events and business growth · Author of 3 ISBN-registered books: «Triunfar o morir en el intento» (2013) and «De esclavo a dueño» (2023) · International keynote speaker for the HORECA sector.
Questions owners ask me before opening one
What is a ghost kitchen and how does the business model work?
What is a ghost kitchen and how does the business model work?
A ghost kitchen is a restaurant with no dining room and no walk-in guests: it produces exclusively for delivery, sells through aggregators or a direct channel, and monetises margin per order instead of covers. It works by removing retail rent and front-of-house payroll, but it hands that saving to the platform commission, which reaches 30% wherever no legal cap applies.
What does ghost kitchen mean compared with cloud kitchen or dark kitchen?
What does ghost kitchen mean compared with cloud kitchen or dark kitchen?
Ghost kitchen, cloud kitchen and dark kitchen name the same thing: a kitchen producing only for delivery orders. The difference is where the term came from, not what the operation does. US operators and ghost kitchens for rent listings favour ghost kitchen, Europe leans on dark kitchen, and the operating model behaves identically in both markets.
How does a ghost kitchen work financially with today's commissions?
How does a ghost kitchen work financially with today's commissions?
It clears cash only when blended effective commission drops under 20% and the average ticket holds. With US tickets of USD 20-35 per Lightspeed 2025, an order pays for itself once the direct channel carries at least a third of volume; at 100% aggregator with 27% to 30% commissions, margin disappears in most categories.
How to start a virtual restaurant business without going under in year one?
How to start a virtual restaurant business without going under in year one?
Start with unit economics, not with the build-out: calculate margin per order on the price net of commission, keep food cost at 32% maximum, and do not open unless it clears USD 6 net. Then validate the 25-minute delivery radius, launch one brand with eight to eleven dishes, and push the direct channel from the first week.
How many virtual brands can one station realistically run?
How many virtual brands can one station realistically run?
One, until it sustains 400 monthly orders with fill rate above 85% and ticket time under 18 minutes at peak. A second brand only earns its place when it shares at least 70% of the first brand's ingredients. Multiplying brands before that stretches ticket time, which is precisely what app customers punish.
Ghost kitchen: 2026 data from official sources
Verifiable industry benchmarks from official, non-commercial sources (government, industry associations, market research) - not competitors.
| Metric | Benchmark 2026 | Source |
|---|---|---|
| Operadores de restaurante que usan IA | Más del 25% de los operadores ya usa inteligencia artificial | National Restaurant Association (Restaurant Dive) 2026 |
| Comodidad de operadores con IA | 86% de los operadores se declara cómodo usando IA (2025) | Toast 2025 |
| Casos de uso de IA en restaurantes | Automatización de marketing 28%, insights en tiempo real 27%, optimización de menú 26% (2025) | Toast 2025 |
| Ticket promedio de pedido de delivery EE. UU. | USD 20-35 por pedido en 2025 | Lightspeed 2025 |
| Marcas virtuales como estrategia de expansión | 32% de las estrategias de expansión de restaurantes en 2025 | Technomic (Apicbase) 2025 |
| Mercado de dark kitchens en India | US$ 552 millones (2023), proyectado a US$ 1.523 millones en 2030 (CAGR 15,6%) | Coherent Market Insights (GlobeNewswire) 2024 |
Related content
Ghost kitchen with the Masterestaurant method
Applied in +8.400 restaurants across 43 countries.