Virtual restaurant business model: why the unit economics are settled before the first sale

Chapter 1 — Marketplace commission is cost of goods sold, not advertising

A virtual brand living on a marketplace pays between 18% and 30% commission per order, against the 3% to 4% a physical dining room pays for card processing, and that gap belongs in the plate cost sheet, not in the marketing line. Book the fee as advertising and the gross margin on your P&L looks healthy while cash drains month after month, and that mismatch between paper and bank explains most of the closures I see in hidden kitchens. The arithmetic is unforgiving: at 32% food cost, the MAXIMUM ceiling under the Masterestaurant framework, with a 25% commission, the operator keeps 43 points for packaging, last mile, kitchen labor and rent. The National Restaurant Association puts optimal food cost between 28% and 35%, but that range was built for a business collecting 100% of the ticket. Diego F. Parra pushes the fee down into plate-by-plate costing, so the list price is born with it already inside.

Chapter 2 — The asset is not the kitchen, it is captured local demand

A 42-square-meter hidden kitchen can be replaced in six weeks with a lease and a combi oven; a base of 4,000 customers with order history, address and frequency never can. Delivery Hero closed 2024 with EUR 48.8 billion in group GMV, up 8% (Delivery Hero, Q4 and FY 2024 Results), and iFood reports more than 380,000 partner establishments across more than 1,500 Brazilian cities (iFood, 2024). Read those two numbers together and the flaw in the model shows up: demand is enormous and it is RENTED. You do not own customers, you hold temporary access to somebody else's customers, and that access gets repriced each quarter to suit the platform. An operator who builds no owned channel — an ordering site, a WhatsApp base, a repeat-purchase program — is paying 25 margin points for an asset that never reaches the balance sheet. The band matters more than the store count, because marketplace commission is negotiated with volume and fixed structure dilutes at very different speeds.

Chapter 3 — Unit economics by revenue band: one model, five different businesses

Below 500 thousand USD a year the operator pays list rate, 25% to 30%, with zero leverage, and break-even demands a high average ticket: at 28% commission and 30% food cost, a 12 USD order leaves under 5 USD of contribution. Between 500 thousand and 1 million the first negotiation appears and the rate drifts toward 20%. Above 1 million an owned kitchen with direct couriers starts to pay for part of the volume. Over 5 million the dominant cost stops being commission and becomes corporate structure. And above 10 million the business buys technology: White Castle passed 100 voice-AI drive-thrus by late 2024, at a 90% order completion rate and roughly 60 seconds per order (Restaurant Dive, 2024). Above 5 million in annual revenue, the media-chef virtual restaurant or licensed brand plays with costs the small band never meets: brand royalty between 4% and 8% of gross sales, centralized quality control, recipe audits in every host kitchen, and a content budget that does not drop below six figures a year.

Chapter 4 — The high end: celebrity brands and large formats above 5 million

Leverage is real, since the brand cuts the acquisition cost platforms charge a stranger, and so is fragility: one scandal involving the face on the label contaminates all twenty kitchens at once. Here technology finally earns its keep. Wendy's announced FreshAI across 500 to 600 US locations by the end of 2025 (CNBC, 2024), and Starship completed 5.8 million autonomous deliveries in 2024 (Forbes, 2025). Those figures only close with volume; at 400 thousand USD a year neither one pays for its own installation. Take a virtual brand with an 18 USD average ticket, 30% food cost and 25% commission, then raise inputs 5%: food cost moves to 31.5%, contribution per order slips from 8.10 to 7.83 USD, and nobody notices. At 12% the food cost hits 33.6%, already past the 32% ceiling, and contribution falls to 7.45 USD. At 20%, food cost of 36%, contribution of 7.02 USD, which is 13% below the base case; a business running at 6% EBITDA has just crossed into losses.

Chapter 5 — What happens if inputs rise 20% and you leave the menu alone

The trouble does not stop there, because the instinctive fix — a 10% price bump on the platform — collides with demand that compares: Deliveroo logged a record 3.5 orders per consumer per month across the UK and Ireland (Deliveroo plc, 2024), and that frequency belongs to the channel, not to your brand. The way out is menu engineering, not a price list. Before building a fleet, measure against the real market cost: Uber Eats couriers averaged 14.96 USD per hour in 2024, down 5%, and DoorDash couriers 12.23 USD per hour, down 3% (Gridwise, 2024). That number is the floor any in-house operation competes against, and once you add benefits, the bike, fuel, insurance and idle time, a payrolled courier rarely lands below it. Which is why owned delivery almost never wins on cost: it wins on DATA and on control of the experience, a different argument that deserves to be said out loud in the board meeting.

Chapter 6 — Your own last mile competes against a courier earning under 15 dollars an hour

The crossover shows up when order density per square kilometer supports three drops per run. Below that, paying commission beats paying idle time, however painful the line looks on the P&L. A virtual restaurant dashboard holds up on four numbers, and gross revenue is none of them. First, contribution per order NET of commission and packaging, in dollars rather than percent, because the percentage hides the ticket. Second, food cost per brand, never consolidated: three brands sharing one oven can average 30% while one runs at 38%. Third, share of sales through owned channels, which should travel from 0% to 20% in the first year. Fourth, acquisition cost per channel against first-order contribution. Cash flow, not profit, is the leading cause of financial stress and closure among small businesses (Inc.), and these four indicators flag it weeks ahead. Start this week with the second one: split food cost by virtual brand and find out which of the three is subsidizing the other two.

Chapter 7 — The four differences that decide EBITDA

COMMISSION IS NOT MARKETING, IT IS COST OF GOODS SOLD. A physical restaurant pays 3% to 4% in payment processing; a virtual brand living on a marketplace pays 18% to 30% for the same order. Filing it under advertising makes gross margin look healthy on the P&L while cash drains. The Masterestaurant framework pushes that fee down into recipe costing, dish by dish, so the list price is born with it inside. THE ASSET IS NOT THE KITCHEN, IT IS CAPTURED LOCAL DEMAND. A 40-square-metre ghost kitchen can be replaced in six weeks; a base of 4,000 customers with ordering history cannot. Delivery Hero closed 2024 with 48.8 billion euros in group GMV (Delivery Hero, FY 2024), and that demand is rented, never owned. An operator who has not shifted at least a third of revenue to direct channels by month twelve has been building equity for the marketplace.

Chapter 8 — The four differences that decide EBITDA — in practice

LAST-MILE COST IS STRUCTURAL AND UNDER PRESSURE. Gridwise measured Uber Eats courier earnings falling 5% in 2024 to 14.96 USD per hour, with DoorDash down 3% to 12.23 USD (Gridwise, 2024). When courier income compresses, the system pushes cost upward through fees and downward through wait times. A model that only clears today's rate has an expiry date printed on it. LOCAL VISIBILITY GETS BOUGHT TWICE OR NOT AT ALL. A ghost kitchen has no window and no foot traffic, so its only storefront is an algorithm: marketplace ranking on one side, the local map on the other. Fight only the first and you stay tied to the platform's auction. Build a verified listing, reviews and polygon-level spend as well, and you earn a second channel whose marginal cost falls over time instead of climbing with competition.