Physical restaurant or dark kitchen: which one wins, measured in unit economics rather than instinct

Which leaves more margin per sale, the dining room or the ghost kitchen?

The dining room leaves more margin per sale: it holds 62% to 70% contribution margin against 38% to 45% for a dark kitchen living inside a platform.

One line of the income statement explains the gap, the channel commission, which per Nation's Restaurant News runs 15% to 30% nominal and reaches 30% or 45% effective once you add forced promotions, packaging and the discount the app charges for visibility. On a USD 15,000 monthly operation, that 24-point spread is USD 43,200 a year the dining format keeps and the ghost kitchen hands over. Now the honest concession: a dark kitchen reaches that number with startup CAPEX three to five times smaller, so you are comparing speed of entry against profitability per dollar sold. The dining room wins, because margin gets collected every month and CAPEX gets paid once. Whoever prices the channel decides the business, and there the physical restaurant wins for a contractual reason, not an operational one.

Occupancy cost: fixed rent versus renegotiable commission

Rent on 120 square meters weighs 8% to 12% of sales and stays locked by a term contract with a known escalation clause; the Rappi or iFood commission gets renegotiated without your signature, and a two-point move on USD 15,000 monthly takes USD 3,600 a year, which is exactly one good month of profit. Look at the scale of your counterpart: iFood closed 2024 with 55 million active customers, Rappi operates in 9 countries and 350 cities with over 500,000 registered partners, and Uber Eats passed one million partner merchants. You do not negotiate with that size. Rent is a HIGH and predictable cost; commission is a variable cost somebody else moves. Choose the cost you can budget. Measured on equivalent sales, the ghost kitchen employs 2.0 people per USD 100,000 annually against 4.1 for the format with a dining room, and that gap matters more than an investment dashboard usually registers.

Jobs created per USD 100,000 in sales

The roles that vanish — server, host, bar, closing shift — are precisely the lowest-barrier ones: first job, labor reinsertion, youth employment. A 45-square-meter dark kitchen billing the same as a 120-square-meter restaurant employs roughly half the people. That is capital efficiency and, at the same time, destruction of formal employment in the segment where it is hardest to create, so the decision is not neutral against SDG 8. For a credit officer with an MSME portfolio, the dining format finances more employability per dollar placed. The dining room wins there, and not out of nostalgia. The physical restaurant generates its own data and the pure dark kitchen generates it for the platform, which decides who holds negotiating power three years out. Reservations, table turns, average check by daypart and guest history are operator assets when the sale runs through your register; when it runs through Rappi, the phone number, the frequency and the preference belong to Rappi.

Who keeps the guest data?

Put plainly, the ghost kitchen operator sells food and buys traffic, and when the platform raises commission two points there is nobody to pass the hit to.

According to Daniel Isenberg, professor at Babson College and creator of the entrepreneurial ecosystems framework, an MSME's fragility rarely sits in the product: it sits in channel concentration. With 28% of restaurants using AI to automate marketing per Toast 2025, whoever lacks an owned database also lacks anything to feed it. A Peruvian food operation in Bogotá billed USD 15,000 monthly with 88% of sales inside two apps, and its consolidated contribution margin sat at 41%. Effective commission, promotions and packaging included, weighed 34 points. After opening a 14-table dining room and keeping delivery as a disciplined channel, the mix settled at 55% dining and 45% platform; consolidated margin climbed to 58% and the dining average check came out 31% above the app check, because at the table you sell drinks and dessert.

A mini-case with cash: the USD 43,200 that changed pockets

Masterestaurant S.A.S., technology partner of SATE Institute, walked through the dish-by-dish costing and the menu was rebuilt with a 32% food cost ceiling. Those 17 margin points on USD 180,000 annually are USD 30,600 recovered in year one, with a new USD 1,400 monthly rent already subtracted. A brief digression fits here, because it touches the same nerve as the whole comparison. Whenever an operation opens a digital channel, the temptation to pull the printed menu and leave only the QR shows up almost every time, and it is the same family of error as closing the dining room to live off the app. The printed menu controls service pace, menu narrative and suggestive selling; the QR solves price updates, accessibility and consultation analytics. Masterestaurant sustains BOTH channels, each with its role, and none of the operations that migrated to QR-only recovered their average check without printing again.

Two channels, two roles: the printed menu and the QR digression

The parallel with delivery is exact: the platform gives reach, the dining room gives margin and relationship. Back to the central point: a channel you do not control is not a channel, it is a lease with an open term and someone else's price. Assume the market behaves exactly as the projections say: Grand View Research estimates a 12.6% CAGR for cloud kitchens between 2026 and 2033, Global Growth Insights calculates 12.7% for dark kitchens between 2025 and 2033, and Precedence Research projects USD 248.10 billion by 2035. What happens to your operation then? Competition for the same square meter of app multiplies, visibility commission rises, and the cost of appearing first stops being marketing and becomes cost of sales. A market growing at 12.6% with a low entry barrier does not distribute margin: it compresses it. It already happened with drive-thru, which fell from 83% of QSR orders in 2020 to 65% in 2025 per Intouch Insight.

What if the cloud kitchen market grows as projected?

Channel growth is not your growth if you do not set the price inside it.

If you hold nine months of working capital, a brand with its own repeat purchase and a location where commercial rent weighs under 12% of projected sales, open the dining room and treat delivery as a second channel capped at 40% of the mix. That is the recommendation for the dominant MSME profile across Latin America and the Caribbean. The pure dark kitchen suits three concrete cases and no others: testing a new concept before committing to rent, absorbing spare capacity in a kitchen that already exists and is paid for, or entering a city where the commercial square meter exceeds what the average check allows. Outside that, it is a bet that channel pricing will not rise, and it rises. Start this week with the number that decides everything: calculate contribution margin per dish and per channel, effective commission included.

Where the two models genuinely split?

The hard difference is not CAPEX, which is what everyone looks at, but who prices the channel. In the physical restaurant the owner sets menu price and occupancy cost is locked by a term lease;

in the pure dark kitchen commission gets renegotiated without your signature, and a two-point move on an operation billing USD 15,000 monthly takes USD 3,600 a year, which happens to be one good month of profit. The second cut is labour, and it matters to any employability programme. On equivalent sales, the ghost kitchen employs 2.0 people per USD 100,000 annually against 4.1 for the floor format, and the vanishing roles —server, host, bar— are precisely the lowest-barrier, highest-turnover first jobs. A credit portfolio financing only dark kitchens optimises return per peso deployed and degrades its own formal-employment indicator. Then comes the data question, which decides three-year resilience.

Where the two models genuinely split — in practice?

Masterestaurant S.A.S. models this with the Restaurant Model Canvas and the result holds up: an operation that knows the name, frequency and favourite dish of its 400 recurring guests survives a platform traffic drop without closing;

one that only sees anonymous orders does not. You can run a dark kitchen with owned data, certainly, but that demands a direct ordering channel, and at that point it is a hybrid rather than a pure dark kitchen. There is a fourth item almost nobody costs: packaging. On the floor the dish travels in china and comes back; in delivery each order burns USD 0.60 to USD 1.40 of material discarded the same day, which on 900 monthly orders reaches USD 1,260 of pure variable cost and a waste volume that collides with SDG target 12.3. Diego F. Parra insists on costing packaging inside the delivery dish rather than in overhead, because hidden there nobody watches it grow.

Where the two models genuinely split — key points?

Finally, the exit horizon. A physical restaurant with a local brand, a guest base and a live lease sells as a going concern at 2.5 to 4 times annual EBITDA;

a dark kitchen with neither brand nor data sells as used equipment. Anyone investing with future liquidity in mind should read that line twice.