Cost stress scenario simulation in restaurants: myth vs reality

Step 1: set your break-even from real purchase invoices, not template percentages

Before you stress anything you need to know the sales figure at which you neither win nor lose, and that number comes from two months of supplier invoices, full payroll including benefits, rent and utilities, never from a 30% food cost copied off a blog. Take your monthly fixed costs, divide them by the average contribution margin per cover —menu price minus raw material cost, dish by dish— and you get the covers you need each month to break even. An independent restaurant with 42 tables and USD 18,000 in monthly fixed costs running a contribution margin of USD 7.20 per cover needs 2,500 covers a month, roughly 83 a day. The deliverable is one cell holding that number plus the cut-off date; you verify it by reconciling period purchases against closing inventory. Without that anchor, everything downstream is astrology done in a spreadsheet. The variables worth stressing are the ones that change without asking permission: protein price, minimum wage, aggregator commission and energy tariffs.

Step 2: name the four variables that actually move on your street

Everything else is dashboard noise. Protein carries between 38% and 55% of raw material cost on most Latin American menus, so a 12% jump in beef moves total food cost a good three points. Delivery commission deserves separate treatment because it is not an innocent variable cost: it is a channel with its own margin, and the dining room may be quietly subsidising it. In Colombia, where 95% of the restaurant market consists of independent operators according to Acodrés, that channel mix decides who survives a bad quarter. Write down the four variables with their current percentage weight over sales; that is the deliverable, and you verify it with addition: the four weights plus fixed costs plus margin have to reach 100%. The scenario worth modelling is not the worst case imaginable but the most likely of the bad ones, because a model built on absurd assumptions gets dismissed and never reopened.

Step 3: build three scenarios, and keep the adverse one plausible rather than apocalyptic

I work with three columns: base —what happens today—, moderate stress and combined stress. Under moderate stress move a single variable: protein +12%. Under combined stress move two or three at once, which is how reality actually arrives: protein +12%, wages +9%, dining room traffic −7%. That last point is the one almost nobody models and the one that kills, since a drop in covers reduces fixed costs by exactly nothing. The deliverable is three columns sharing the same row format plus a closing row reading «covers required for break-even» in each. You verify it by checking that no column rests on an assumption you cannot back with an invoice or a recent news headline. A model that does not end in a dated decision is not a model, it is a hobby. For each scenario write what you do and when: if combined stress demands 104 covers a day and you run 83, you carry a 21-cover gap that closes by pulling the four dishes with below-average contribution margin, cutting one weekday support shift and raising three anchor prices, not twelve.

Step 4: turn every scenario into a dated decision, not a pretty chart

Diego F. Parra keeps insisting that the menu gets pruned BEFORE the cash shortfall makes the call, because whoever cuts with a bank account in the red cuts badly, and cuts people instead of dishes. The Masterestaurant methodology ties every scenario to an operating action with an owner and an execution week. The deliverable is a three-row table —scenario, decision, week— taped up in the office; you verify it by asking the chef which dish sits first on the pruning list. The model does not have to be pretty, it has to be weekly, and that is the whole difference between an exercise that saves a business and one that decorates a folder. Forty-five minutes every Monday with last week's purchases, covers by service band and real payroll are enough to refresh the three columns. The reasoning is arithmetic: when a cost slides three points, monthly review tells you once you have already burned twelve weeks of margin, while weekly review hands you eleven weeks of lead time to renegotiate, reformulate a recipe or move a price.

Step 5: make the exercise weekly, even when weekly hurts

The US restaurant sector employs 15.7 million people in 2026 and projects 17.3 million by 2036 according to the National Restaurant Association; that growth is not carried by operators who look at their numbers in December. Deliverable: a file with the date in its name and five consecutive versions. One version only means the system does not exist. The most expensive mistake is loading payroll and rent into plate cost, because it inflates apparent food cost, pushes prices up where they should not move and leaves break-even invisible. Payroll, rent and utilities belong to break-even, not to the plate; the ceiling for raw material per dish is 32% and that ceiling is a maximum, never a target. Second mistake: stressing with round percentages nobody can back —a 20% rise across the board has never happened and produces a model everyone ignores. Third, modelling costs alone while traffic sits frozen, when falling covers are the most violent variable in the system.

The mistakes that sink the exercise, and how to dodge them

Fourth, keeping it in a file only the owner opens. Fifth, mistaking a budget for a stress test: a budget projects what you expect, a simulation projects what happens when your expectations fail all at once. Fix those five and the model starts paying for itself in month one. Picture a restaurant running 30% food cost today with a contribution margin of USD 7.20 per cover. Protein climbs 12%, wages 9%, and traffic gives up 7% because of roadworks on the block. Food cost reaches 35%, margin per cover falls to USD 5.60, and break-even jumps from 83 to 107 daily covers while actual sales slide to 77. That is a thirty-cover daily deficit which supplier credit papers over for five or six weeks, until the supplier stops extending terms and the full hole becomes visible. That restaurant does not close because beef got expensive: it closes because nobody measured those three points against break-even while daily cash still felt comfortable.

What would happen if you skip it: the arithmetic of closing down?

And the bill is not only the owner's; US foodservice surplus food hit USD 157 billion in 2024, some 14% of its sales according to ReFED, and that money leaks away restaurant by restaurant, unmodelled.

You know the exercise landed when you can answer five questions without opening the file. How many covers you need today to avoid losing money. How many you would need if protein rose 12%. Which dish leaves the menu first and which one second. What week you execute the cut if the combined scenario turns real. And who updates the file every Monday when you are not around. Fail one of them and the model is incomplete, and the one that fails most is the fifth, because the owner keeps everything in his head and on his laptop. Hard verification: ask the chef for the break-even cover count and ask the manager for the date of the last update; if both answers match the file, the system is alive.

Closing checklist: how you know the job was done right

If either one drifts, what you have is a document rather than a management tool, and that gap gets paid in cash. The myth says these exercises belong to chains with a finance department. Reality: an independent venue with a spreadsheet, two months of real purchase invoices and covers counted by time slot reaches the same diagnosis as a chain running corporate software, because the model is not the hard part. Clean data is. Second myth: that stressing costs is a supplier-negotiation tool. It serves a less comfortable and far more useful purpose, which is deciding which dishes leave the menu and which shift gets cut BEFORE the cash shortage decides for you. Negotiation comes later, and it goes better once you know your true ceiling. Third myth, the expensive one: that the adverse scenario equals the worst case imaginable. It does not. Model the probable-bad —inputs +12%, occupancy −8%, aggregator commission +2 points, all in the same quarter— because that is what actually happens, while the catastrophic version paralyzes and nobody acts on it.

Where the myth breaks?

There is a genuine tension in this craft worth resolving head-on: the more granular the model, the fewer people maintain it; the simpler it gets, the coarser the diagnosis.

The way out is not a lukewarm middle ground but a hierarchy: four variables stressed weekly with hard data beat twenty variables updated whenever somebody remembers. Start with four. Granularity is earned later, once the habit exists. On menus and QR codes, which always surface here because prices change: the house recommendation is keeping the PHYSICAL menu and adding the QR menu as a complement. The physical menu governs service pace, menu narrative and suggestive selling; the QR handles delivery, accessibility, price updates and analytics. Both, each in its role. Killing the printed menu to save on printing costs more in average ticket than it ever saves in paper.