Food cost: the mistakes draining your till and the method that actually works

Price the standard recipe before you touch the calculator

The first deliverable is not a percentage but a per-dish spec sheet with exact portion weight, declared trim loss and unit price taken from the latest supplier invoice, not from the chef's memory. Write every ingredient in the unit you buy it and the unit you serve it, with its yield factor: a loin arriving at 4.2 kilos that leaves 3.1 kilos portioned yields 73.8%, and that 26.2% of trim belongs to the plate cost even though nobody watches it leave the storeroom. It is done when every menu item has a signed spec sheet and the sum of its ingredients matches, gram for gram, what the cook puts on the pass. Verify it by weighing three plates at random during service; if the real portion strays more than 5% from the sheet, the sheet does not exist yet, only a wish does.

Freezing your purchase price is 2026's most expensive mistake

Reprice your spec sheets every time an input moves more than 5%, because this year the arithmetic runs against you: USDA ERS projects +3.2% for all food, +3.6% for food away from home and a brutal +7.5% for beef, with the U.S. cattle herd at a 75-year low (Food Price Outlook, June 2026). Coffee and nonalcoholic beverages climb 5.7% per the same source. A protein dish costed in January and left untouched until December will have shed four to six margin points without anyone changing a single recipe. Set up an automatic recalculation the day the invoice arrives, not the day of the accounting close. The deliverable is a log with date, input, old price, new price and dishes affected; if that log sits empty for a whole month, somebody stopped looking. Theoretical cost comes from multiplying units sold by each dish's spec-sheet cost; actual cost comes from opening inventory plus purchases minus closing inventory.

Theoretical cost against actual cost: the only subtraction that makes money

Subtract them and you get the gap, and that gap, in most kitchens Masterestaurant audits, sits between 3 and 6 points. A 240-cover kitchen in Guadalajara billed 1.9 million pesos a month believing it ran a 28% food cost; once the recipes were priced against that same month's invoices, the real figure was 34.7%. Nobody was stealing. Those 6.7 points lived in undeducted prep waste, portions served by eye above spec, and eight dishes whose price had been frozen for fourteen months. On that revenue it amounts to 127,000 pesos a month walking out of the till without a single named line on the income statement. Split the difference into four buckets and put a name against each one: portioning, prep waste, theft or leakage, and menu-price error. Weigh the hot line's garbage five days running and you will have real waste in kilos rather than hunches; ReFED documents that every dollar invested in preventing food waste returns seven dollars in future benefit, a 600% ROI no software vendor will match.

Chase the gap by family, not by scattered dishes

Portioning gets fixed with calibrated ladles and a scale on the line, never with memos. Whatever fails to close after those four buckets, and it is usually under one point, is your tolerable margin of error. It is done when you can explain, with a figure attached, where every point of last period's difference came from. Rank the menu by absolute contribution margin multiplied by units sold, and only then look at the percentage. A dish at 34% food cost leaving 210 pesos per sale beats one at 22% leaving 61, and menus edited by percentage have usually executed their best seller without noticing. I got this wrong for years, and I will say it plainly: I believed food cost was a monthly number, the one you get dividing purchases by sales, and that number is good for sleeping well and nothing else, because it averages the dishes feeding you with the dishes eating you.

The percentage lies; the cash decides who stays on the menu

With full-service margins running 3% to 8% per WhippleWood CPAs' 2026 benchmarks, getting that order wrong costs you the whole year. Add direct inputs to total labor —wages, benefits, overtime, turnover— and measure that prime cost against sales, because 28% food with 41% payroll leaves you exactly as ruined as 38% food with 28% labor. Cut first where the number hurts: if your rent already weighs what it weighs —around 53 dollars per square foot per year in Los Angeles according to Pepperlot for 2025— prime cost is the only lever you move week to week. Cornell measured that roughly 26% of new restaurants close or change hands in year one and 60% never reach three years, and the cause is rarely the food: it is a structure costed exactly once. The deliverable is a weekly board with three figures: food, labor, sum.

Common errors that wreck a well-intentioned costing

Loading payroll, rent and utilities onto plate cost is the most repeated and most expensive mistake, because it inflates food cost artificially, pushes you to raise prices where you did not need to, and hides the real problem, which is your break-even point. Another classic: costing at list price instead of the price actually paid after discounts and returns. You also see closing inventory counted from memory on a Sunday at eleven at night, which corrupts the whole period's actual cost. And the subtle trap, a food cost under 32% achieved by shrinking portions: you saved nothing, you traded today's margin for the customers you lose in six months. Mexico has more than 641,000 restaurants per CANIRAC and INEGI (2024); the survivors are not the ones cutting portions, they are the ones counting properly. Walk this list on the last working day of the month and refuse any yes without a figure beside it.

How to know it all landed: the close that verifies itself?

Spec sheets: 100% of active dishes priced, with a last-update date inside 30 days. Inventory: counted by two people, in a format that respects the purchase unit, closed before the first delivery of the following month.

Gap: theoretical against actual, calculated and explained point by point, aiming to push it under 2 points by the third month of discipline. Prime cost: logged on the same board, with its twelve-week series. Menu prices: your eight highest-volume dishes checked against today's input cost. If a single row lacks a number, the close is not done. Start tomorrow by weighing the hot station's waste for five days. Bad costing answers "what did I spend"; good costing answers "what SHOULD I have spent and why the mismatch". Only that second question produces money, because it turns an accounting figure into a list of actions on portions, receiving and waste.

Four differences between useful costing and a decorative exercise

Percentage misleads; currency decides. A dish at 34% food cost leaving 210 pesos of contribution margin beats one at 22% leaving 61, and menus edited on percentage have quietly killed their best earner more than once. Food cost alone says nothing about viability: prime cost rules, direct input plus total labor together, because 28% food with 41% payroll ruins you exactly as fast as 38% food with 28% labor. Correct costing is a dated cycle: inventory at the same hour, recosting when an input moves 5%, quarterly menu review. Without a calendar there is no control, only a spreadsheet aging on the owner's desk.