Restaurant business model: what the numbers say and what the myth keeps repeating

Prime cost is the exam your model either passes or fails

A restaurant business model is validated when prime cost —food plus fully loaded labor— stays under 62% of net sales for eight of twelve months, not when there is a line at the door on Friday. That Bogotá owner was billing 412 million pesos a month, 19% more than the prior year, and still could not cover payroll: his prime cost had climbed from 61% to 69% in fourteen months, eight points that on those sales mean 33 million pesos a month gone without anyone signing an approval. Gross sales went up and covered the hole. If you track one number this week, track that one: the daily sales figure on the POS screen is the most addictive and least informative data point in this industry, because it describes activity, not survival. Growth hurts cash when contribution margin per dollar sold falls faster than volume rises, and that almost always comes down to channel mix.

Why can growing 19% make cash worse?

Run the counterfactual all the way: you add 20% in new sales and all of it arrives through a marketplace, where commission takes between 18% and 30% of the ticket;

a dish that returned 68% contribution in the dining room drops to 44%, so 100 new pesos deliver 44 instead of 68. Same oven, same kitchen, more people shouting tickets, and you needed 24 extra pesos of contribution just to break even. You do not have them. The market pushes that way: iFood holds 40% of active delivery users in Latin America and 89% in Brazil, while in Mexico DiDi Food and Rappi add up to 74% (Sensor Tower, 2025). The question is not whether to enter, it is at what price and with which menu. Sector growth in 2026 will not fix anyone's model, and that is the figure that settles a board meeting fastest. The National Restaurant Association projects real growth for the United States, inflation already stripped out, of just 1,3% in its 2026 State of the Restaurant Industry.

The sector barely grows, so margin has to come from inside

Brazil closed 2025 at 0,92% real over twelve months, according to Abrasel. Mexico looks better, near 6% nominal per CANIRAC, but nominal and inflation are cousins, not twins. Translated into decisions: if your 2026 plan depends on the market handing you volume, your plan is a bet. Between one point and six of market growth there is no room for an 8% to 15% operating margin; that margin is manufactured inside the building, pulling food cost from 35% down to 30% and reordering channel mix, not waiting for the neighborhood to eat more. Thirty-two percent food cost per dish is the MAXIMUM you tolerate, never the goal, and mixing those two up costs margin points every single month. In a clean model the plate carries raw material and nothing else: payroll, rent and utilities are not allocated per dish, they belong to break-even, because the day you push rent into the cost of a steak you stop knowing which dish pays your bills.

Food cost of 28-32% is a ceiling, not a target

Diego F. Parra insists at Masterestaurant on reviewing first the six dishes that move 60% of units sold, which is where a three-point swing turns into real money: in an operation billing 400 million a month, three points of food cost are 12 million. Menu engineering is not a menu-design exercise, it is a cash decision with a first and last name attached to every reference. Benchmarks read differently by size, and applying them flat is the most expensive mistake on this list. Small operation, one location, up to 80 million in monthly sales: the owner counts inventory by hand every week, prime cost up to 65% is livable because he covers shifts himself, and a realistic margin runs 6% to 10%. Mid-size, two to five locations: a manager per store appears, prime cost has to fall to 60-62% because administrative payroll now weighs, and food cost closes at 30%.

How to read these numbers in YOUR operation?

Group, six locations or more: the central office shows up and usually eats 4% to 7% of consolidated sales, so prime cost must live at 58-60% for EBITDA to hold 12-15%.

A single store at 60% prime cost is healthy; a group at 60% is walking a tightrope. These numbers come from public sector sources, and it is worth stating what they are NOT. Growth figures come from the National Restaurant Association (2026 State of the Restaurant Industry, 1,3% real in the U.S.), from Abrasel (0,92% real in Brazil, 2025), from CANIRAC (near 6% in Mexico, 2025), and from Sensor Tower for delivery share across Latin America; the global delivery market split comes from Towards F&B, which puts Asia-Pacific at 34%, North America at 31% and Europe at 27% for 2025. They are averages of large markets with inflation rates and labor structures unlike yours, and none of them replaces your own closed P&L.

Where these benchmarks come from and how far they reach?

Treat them as direction, not verdict: if your prime cost sits eight points above the range, the problem is yours, not the benchmark's, and Monday's inventory will show it.

What separates a validated model from one that merely looks healthy is measurement frequency, and it is paid in margin points. The serious operator closes inventory weekly and calculates prime cost off that close; the other waits for the accountant's P&L, which arrives sixty days late, by which time the food cost variance has already turned into supplier debt and an overdraft. Sixty days of blindness on 400 million in monthly sales, with a variance of just two points, is 16 million you cannot claw back because the product was already served and eaten. For years I recommended closing every two weeks to save admin work, and I was wrong; a fortnightly close hides a bad week inside a good one.

Measure on Monday or find out in March

Weekly, with a physical count of the twenty references that make up 80% of purchasing spend. A restaurant investor asks two questions before signing, and neither one is about Friday's full house: how much of every dollar sold survives to EBITDA, and how many consecutive months it has done so. A 12% operating margin across nine months is worth more than 20% in one quarter and 4% in the other three, because the first describes a system and the second describes seasonal luck. Sector employment backs that reading: in Brazil, every 1.000 direct jobs in bars and restaurants generate 2.250 indirect ones (Abrasel, 2025), confirming we work in a people-intensive business where payroll is not optimized by cutting heads but by output per hour. Concrete action: open your last twelve months of P&L, calculate prime cost month by month, and count how many months came in under 62%.

The investor's two questions, and this week's action

That count is your business model. MEASUREMENT. A validated model calculates prime cost weekly against a closed inventory; the other one reviews the P&L when the accountant shows up, sixty days late, once the food cost variance has already turned into supplier debt. Measuring on Monday instead of in March is worth 8 to 11 points of annual margin, according to the Restaurant365 benchmark for 2026. REVENUE STRUCTURE. A restaurant with real financial maturity knows what share of sales comes from the dining room, first-party delivery, marketplace and catering, and it knows the contribution margin of each one separately. The one that does not know believes it is growing while it shifts sales from 68% contribution toward channels at 44%, with the same kitchen and more stress. COSTING. In a disciplined model the dish carries raw material, real waste and packaging; that is all. Labor, rent and utilities live at the break-even line, where they can be negotiated and sized.

Four differences between a validated model and one that merely sells

When someone spreads payroll across dishes to "know what each one costs", the resulting number moves with volume and helps decide nothing. VALIDATION. A serious restaurant investor never asks how much the location sells; they ask how many consecutive months it has posted positive EBITDA and how much working capital it needs to survive a slow month. Those two answers beat any five-year projection built on a spreadsheet with 15% annual growth typed in by hand.