Restaurant business model: 18 numbers from 2026 that separate the myth from reality

How big is the market you are actually selling to?

Market size is no longer the scarce variable: frequency is, and that is where your model holds or collapses.

In the United States, 77.3% of consumers eat out at least once a week, according to Restroworks, and eating away from home took roughly 39% of total household food spending in 2024, per the American Farm Bureau Federation. Demand is plentiful. Tolerance for wasting that demand is not. With an average check of USD 54 in 2024 against USD 48 in 2023 (US Foods, via Escoffier), the guest pays 12.5% more for the same outing and expects it to be worth the money. These three figures together trigger a single decision: stop chasing new customers and calculate how many times a month the ones already inside come back. Define the channel before the concept, because the channel fixes payroll, rent and whatever margin survives. Seventy percent of American diners ordered delivery in the past month, according to Escoffier, and delivery-only kitchens already account for 41% of the global dark kitchen market as of 2024, per Credence Research.

The channel sets your cost structure before the menu does

That number does not say «open a dark kitchen». It says something less comfortable: a large third of the category runs WITHOUT a dining room, without servers, without a public restroom, and competes for the same guest carrying a structure you do not have. If your site pays premium-district rent to serve orders that leave through a side door, you are financing square meters your model never charges for. Renegotiate the lease or renegotiate the channel; not both at once. A guest who returns twice a month is worth more than any campaign, and the 2025 numbers say almost nobody captures that value. Restroworks measured that 55% of diners visit at least twice a month the restaurants where they belong to a loyalty program; Voucherify reports that 52% of QSR customers already belong to one and that 81% of consumers would join if offered. On the other side, William Blair found that 55% of American diners are members of NO program at all.

Loyalty is not a points program, it is what holds up your break-even

There is the gap: eight in ten would say yes and half were never invited. The takeaway is arithmetic, not motivational. If your break-even demands 1,400 covers a month and your recurring base delivers 400, you are buying the rest every single month at advertising prices. At Masterestaurant, the opening question in a business-model audit is never how much you sell but how much each dish leaves after its own food cost. A dish at 30% food cost hands 70 cents of every dollar to covering the structure; one at 38% hands over 62, and those eight points, across 1,400 monthly covers at a USD 20 ticket, are USD 2,240 a month that simply stop existing. Diego F. Parra insists on an order that almost nobody respects: payroll, rent and utilities do NOT get loaded onto the plate, they belong to break-even. Spread rent inside each recipe cost and the price climbs, volume drops, and you end up with less money to pay that same rent.

Contribution margin per dish: the number Diego F. Parra checks first

It is the most expensive trap in the trade because it looks like accounting prudence. Suppose your casual dining loses 4.3% of traffic year over year, which is exactly what Rezku measured in the United States for 2025. On 1,400 covers at a USD 20 ticket, that is 60 covers fewer, USD 1,200 of lost sales per month. If your average contribution margin runs at 65%, USD 780 of contribution stops arriving; and since base rent and payroll did not drop a cent, that hole comes straight out of profit. Repeat it three months and you are at USD 2,340, more than a full kitchen payroll costs in many markets across the region. Technomic reported the independent restaurant sector shrank 2.3% in 2025, a net loss of more than 9,500 locations. They did not close for lack of customers. They closed because their structure never learned to shrink.

Your value proposition has to survive a stranger's verification

«Home cooking with real flavor» is not a value proposition: it is the owner's wish painted on a wall. The usable version sounds different, and it can be audited: the only place in the neighborhood where a family of four eats a full lunch for under USD 30 in less than 40 minutes. There is price, there is time, there is a consumption unit and there is territory, four variables a stranger can time and compare. With the national average check at USD 54 (US Foods, via Escoffier), whoever solves four covers below USD 30 is playing a different game and can prove it. The pandemic left the memory of what happens when context shifts: Mexico's restaurant industry GDP fell 29.3% in 2020 against 2019, according to INEGI and CANIRAC. Survivors had a concrete promise their guests could repeat from memory. An investor sent me a 42-page business plan last year with five-year projections and exactly one cost figure: «target food cost 30%».

The 42-page plan and the line that was missing

Not one payroll line per shift, not one rent-against-sales line, no average-ticket assumption tested against the neighborhood. That document described a restaurant that did not exist yet, and described it from the outside. A business model actually answers four questions almost nobody writes down: who am I selling to, what do I solve that others do not, what does it cost me to deliver it, and how many times a day can I repeat it without breaking the kitchen. Concept, decor and name are consequences. Invert that order and the restaurant is born with a cost structure no amount of later commercial effort can correct, however good the marketing gets. Prime cost below 60% of sales: measure it weekly, never monthly, and if a Monday reads 63%, cut one shift from that week's schedule and renegotiate your most expensive supplier before Friday. Contribution margin per dish above 65%: put the whole menu on one sheet, sort high to low, and the bottom four dishes either raise their price, change their recipe or leave this month.

The 3 numbers you should tattoo on yourself

Break-even reached before the 20th: divide fixed costs by your average contribution margin, get the covers you need, and mark them on the calendar; if the 20th finds you at 70%, you have ten days to close thirty percent, and discounts will not do it — the 55% who already return twice a month will (Restroworks). Start Monday with prime cost. FIRST: a profitable model sets price from contribution margin per dish, never from total allocated cost. Payroll, rent and utilities do not belong in plate cost — they belong in break-even — and that single distinction rewrites the whole menu. A dish at 30% food cost leaves 70 cents of every dollar to cover the structure; push rent inside the plate cost and the price climbs, volume drops, and you end up with less money to pay that very rent. SECOND: the value proposition of a working model can be verified by an outsider.

The five differences that decide whether the model holds

«Homestyle cooking with real flavor» is not a value proposition, it is a wish. «The only place in the neighborhood where a family of four eats a full lunch for under 30 dollars in less than 40 minutes» qualifies, because it can be measured, it can be broken, and it can be defended when a competitor opens on the corner. THIRD: a model that holds carries an explicit repeat number. How many times a month the guest returns, and what that frequency is worth in cash. An 18-dollar ticket visited monthly and a 12-dollar ticket visited weekly are not the same business even with a similar menu: the second bills 2.6 times more per customer per year and leans far less on acquisition marketing. FOURTH: in a healthy model, every channel carries its own menu and its own price.

The five differences that decide whether the model holds — in practice

Delivery at a 25% commission cannot sell the same dish at the same price as the dining room, and the fix is not «absorbing the commission» but redesigning what gets sold there: high-margin items, packaging that does not wreck the product, and off the channel anything that arrives cold or photographs badly. FIFTH: a validated model was tested before the lease was signed. One hundred real tickets in a provisional format — a kitchen rented by the hour, a market stall, three weeks of pop-up — cost a fraction of a five-year contract and answer the only thing that matters: whether someone pays that price for that food in that context. I got this wrong for years, recommending market studies when the question was settled by selling a hundred times.