From −6.1% to +11.4% EBITDA: shared vs own cloud kitchen in a three-brand operation, settled with the Restaurant Model Canvas and Demand Radar

The diagnosis nobody wants to hear: 6% owned demand

Three thousand one hundred monthly orders, and only 186 belonged to the owner — 6% owned demand after twenty-six months running three virtual brands out of a shared hub. The other 2,914 came through aggregators, with an 11.80 USD average ticket and 780 thousand USD in annual revenue that looked healthy on the bank statement and was fragile in structure. Eight employees, two shifts, no dining room, no foot traffic. The man wanted to move into his own kitchen because the hub's variable rent hurt every payday, and that move, made with 6% direct demand, would have swapped a commission that rises and falls with sales for CapEx that forgives no slow month. The business would have died of cash flow by month four, not for lack of customers, but because the customers were never his. Variable rent in a shared cloud kitchen is not occupancy, it is an extra commission stacked on top of the platform's and aimed at exactly the same slice of the ticket.

Why does the hub's variable rent behave like a second commission

When the hub charges a percentage of sales and the aggregator takes another, the best-selling dish — the one holding the mix together — ends up yielding less than the second or third, and the operator misses it because he watches total sales instead of contribution margin per SKU. In this operation, with an 11.80 USD ticket, each percentage point of variable occupancy was worth roughly 366 USD a month across 3,100 orders. Fixed rent in your own kitchen does the opposite: it dilutes with volume and rewards whoever grows. That is the paradox of the shared model, which lowers the risk of entry and makes success expensive. The crossover between shared hub and owned kitchen in this operation sat at 2,400 monthly orders per brand, and below that figure the shared option wins every time. The arithmetic is dry: as long as fixed rent divided by orders beats the variable percentage applied to an 11.80 USD ticket, you are overpaying for square meters that produce no demand for you.

The real crossover point: 2,400 monthly orders per brand

With 3,100 orders split across three brands, none of them individually reached the threshold, and that single number ended the real-estate debate in twenty minutes. The other condition, the one almost nobody measures, was that at least 35% of orders had to arrive through direct channel. At 6%, an owned kitchen was not a cost decision: it was a bet that Rappi's algorithm would stay generous for the thirty-six months of the lease. We applied the MASTERESTAURANT Direct Channel Traffic Light, the tool that classifies every order by who owns the relationship rather than by technical origin, and the operation went from arguing about square meters to arguing about ownership. Red below 15% direct demand, amber between 15% and 34%, green from 35% up: only in green do we authorize CapEx for an owned kitchen. The owner entered red at 6%, moved to amber by month seven at 22%, and reached green at month fourteen with 34%.

How we applied the MASTERESTAURANT Direct Channel Traffic Light?

The lever was surgical:

WhatsApp Business with a catalog per brand, a repurchase coupon printed on the packaging with a unique code per virtual brand, and a Google Business Profile listing that the shared hub, by definition, would never let him register in his own name. First you build the asset. Then you sign the lease. By month eighteen the operation billed 1.04 million USD a year, with 34% of orders through direct channel and 4,180 monthly orders, and only then did we move two of the three brands into an owned kitchen while the third — desserts, the most seasonal — stayed in the shared hub. Aggregate contribution margin rose 9.4 points, and not through operational magic: 5.1 points came from dropping aggregator commission on a third of the volume, and 4.3 from replacing variable rent with fixed rent already diluted across 4,180 orders. Average ticket moved from 11.80 to 13.60 USD, pushed by direct channel, where the customer builds a bigger order because he is not comparing twelve brands in a grid.

The eighteen-month result, measured in cash rather than narrative

Hiring two more kitchen staff paid for itself out of the first quarter's savings. Sector noise pushes operators toward bad decisions, and it pays to look at the numbers before signing anything. The United States runs roughly 7,606 active ghost kitchens according to OysterLink (2025), and a sizable share are virtual brands born and buried within the same quarter. Delivery Hero closed 2024 with group GMV of 48.8 billion euros, up 8% according to Delivery Hero (2024), while Just Eat Takeaway moved 8 billion euros of GTV across northern Europe on barely 4% constant-currency growth according to Just Eat Takeaway.com (2024). Deliveroo reported a record frequency of 3.5 monthly orders per consumer in the United Kingdom and Ireland according to Deliveroo plc (2024). The platforms grow. The operator who depends on them collects what is left, and no bigger kitchen fixes that asymmetry. The recommendation changes by revenue band, and so does this week's first step.

Transferable lessons by annual revenue band

Below 500 thousand USD a year: stay in a shared hub and measure today what share of your orders comes with the customer's own phone number; if you cannot calculate it within an hour, that is your real problem. Between 500 thousand and 1 million, this case's band: launch the repurchase coupon with a per-brand code this week and lock the 35% threshold before you look at a single location. Above 1 million: audit contribution margin per SKU and per channel, because money is already leaking through dishes that yield differently depending on where the order enters. Above 5 million, the media-chef profile with licensed brands: separate the production kitchen from the brand kitchen and negotiate on consolidated volume, never per unit. Above 10 million, multi-site groups: convert two hubs into owned kitchens as a pilot and compare twelve months before touching the rest of the network.

Limits of this case

I would not expect this result in three contexts, and saying so matters more than celebrating the number. First, in a city under 300 thousand inhabitants, where the digital buyer base cannot sustain 2,400 monthly orders per brand and direct channel stalls around 15% by pure population arithmetic; there the shared hub is the destination, not an intermediate rung. Second, in operations with a single virtual brand and a ticket below 8 USD, because an owned kitchen's fixed cost spreads across fewer dishes and less absolute margin per order, which pushes the crossover far beyond anything reachable. Third, in markets where the dominant aggregator penalizes anyone driving direct channel in its ranking: we have seen contracts with parity clauses that make the repurchase coupon brutally expensive. Read your contract before you move a single customer. DEMAND OWNERSHIP. A shared cloud kitchen does not steal customers, yet it does nothing to help you build them, because the hub address is not yours and cannot be ranked in Google Maps under your brand.

Four differences that decide the call, and none of them is the rent

An own kitchen gives you a Google Business Profile with a verifiable address, and that listing is the one local digital asset no algorithm update can confiscate. At 6% direct demand you are a Rappi supplier; at 34% you are a business. OCCUPANCY COST STRUCTURE. Variable hub rent behaves like a second commission, and it stacks on top of the aggregator's until it eats the margin of your best-selling dish. Fixed rent in an own kitchen dilutes with volume. The crossover point in this operation sat at 2,400 monthly orders per brand: below that figure the shared hub won in every scenario we ran. DISPATCH SPEED AND RANKING. Rappi, Uber Eats and DiDi algorithms punish preparation time with fewer impressions, and that punishment compounds. Inside the shared hub, with four operations fighting over the fryer at the 20:00 peak, average dispatch time was 24 minutes. In the own kitchen, with the line designed around three brands and a dedicated fryer, it dropped to 16.

Four differences that decide the call, and none of them is the rent — in practice

That gap moved the chicken brand from sixth to second place in its category within the delivery radius. REVERSIBILITY OF THE MISTAKE. A shared hub contract ends with 30 or 60 days of notice. An own kitchen with civil works already done ends by selling assets at a 40% to 60% haircut, assuming a buyer shows up. When uncertainty about demand runs high, paying more to keep the exit option open is the rational call, even when the spreadsheet insists fixed rent is cheaper. Here sits the paradox of the trade: the structure that costs MORE per order can cost LESS per unit of risk, and confusing the two is what bankrupts dark kitchens from scratch.