How to calculate restaurant food cost: the errors that eat your margin and the method that returns it

Chapter 1 — Why dividing the invoice by the menu price lies to the owner

Dividing your ingredient cost by the menu price gives you a number that does not exist, and that number is usually off by six to nine percentage points. You buy tenderloin at 18 dollars a kilo, divide it into a 64-dollar plate with a 200-gram portion, see 28% and close the spreadsheet. What the spreadsheet ignores: fat and sinew trim, cooking loss running between 18% and 25% on red cuts, the returned portion, the staff discount and the plate that came back. When you close the month with a real inventory count, 36% shows up. On a million dollars in annual sales, those eight points are 80,000 dollars gone. And it is almost never theft: it is a measurement system nobody ever built. The USDA logged 80-90% ground beef at 5.63 dollars per pound in mid-2026 against 4.56 in 2025, so the error compounds every quarter.

Chapter 2 — Theoretical cost comes from the recipe and the yield, never from the invoice

Cost your plates on NET YIELD, not on the price your supplier charged you. If that kilo of tenderloin comes in at 18 dollars and leaves you 680 servable grams after trimming and cooking, the served kilo costs 26.47 dollars — 47% above the invoice. That gap is pure arithmetic, not judgment or intuition. A proper recipe card carries four columns per ingredient: gross quantity, yield factor, net quantity and extended net cost; without the second column everything downstream lies. The average restaurant wastes between 4% and 10% of what it buys according to The Restaurant HQ (2025), and a good part of that range hides inside a yield nobody measured. Start with your ten highest-turnover plates, weigh them raw and cooked for one week, and the lost points will show up with a name attached. The formula your accountant should be demanding is opening inventory plus period purchases, minus closing inventory, divided by sales for THAT same period.

Chapter 3 — Real cost is measured by inventory difference, never by adding invoices

Adding up the month's invoices produces a figure that swings for no reason, because it ignores what stayed in the walk-in and what was consumed from last month. One large oil order on the 28th ruins your January close and gifts you February. With a biweekly physical count on the twenty items that concentrate 80% of spend — proteins, dairy, oils, spirits — variance drops into measurable range within two cycles. Diego F. Parra keeps hammering one uncomfortable point at Masterestaurant: a food cost figure you cannot reproduce with a physical count is not a figure, it is an opinion. ReFED calculated foodservice at 17.9% of the United States food surplus in 2024. Food cost variance is the subtraction between what the recipe says the plate should have cost and what inventory says it did cost, and your target is holding it under 1.5 percentage points. At 30% theoretical and 31.2% actual you are fine.

Chapter 4 — Variance between theoretical and actual is the only KPI that matters

At 30% theoretical against 36% actual you have a portioning, receiving or yield problem, and the good news is that you already know where to look. Variance turns a conversation about suspicions into a list of three or four specific ingredients. The National Restaurant Association reported that 90% of full-service operators raised prices in 2024 and 60% removed menu items; raising prices before closing the variance is patching a hole with the customer's money. Measure your five most expensive items weekly and the rest monthly: frequency kills more variance than any software. One point of food cost means five different things depending on what you bill per year, and that is exactly why generic benchmarks do damage. Below 500,000 dollars, a point is 5,000 dollars and the owner costs by hand in a spreadsheet: what matters there are recipe cards for the ten plates driving 70% of sales.

Chapter 5 — The same food cost point weighs differently depending on your revenue band

Between 500,000 and a million, the point is worth up to 10,000 and the first chef who buys without recipe cards appears. Above a million you need formal biweekly counts. Over five million, a single point is 50,000 dollars a year and control shifts from a task to a job position. Above ten million, with profitable operators running labor at 34.2% against the 36.5% average (National Restaurant Association, 2024 data), food cost variance decides EBITDA for the entire group. Above five million dollars a year, the celebrity-chef restaurant or the large-format themed venue runs a theoretical food cost between 33% and 38%, and that is fine as long as contribution margin per seat holds it up. The problem is not the percentage but the volatility: menus rotating every eight weeks, imported proteins, mise en place carrying three days of labor, and preparation waste that at this level reaches twice what a bistro sees.

Chapter 6 — High end: the celebrity-chef restaurant pays costs nobody budgets for

When Brazil concentrates roughly 38% of global coffee supply according to Bellwether Coffee, a frost there moves your dessert menu here. The discipline that works is recosting the full menu before each card change, not after, and setting a food cost ceiling per plate family instead of a blended average that hides the three dishes bleeding out. Suppose your food cost sits at 31% on 1.2 million dollars in sales and your basket climbs 3 points, hardly exotic when retail eggs rose 21.9% in 2025 per the USDA. Change nothing and you lose 36,000 dollars of gross margin a year, which in most full-service operations equals the entire net profit. Raise prices 4% across the board and you will probably lose traffic on the elastic items. The route that actually works: menu reengineering on your six highest-volume plates, portion adjustment where the guest does not notice, substitution of two imported items for local equivalents, and renegotiation of the three purchases that concentrate spend.

Chapter 7 — What happens if input inflation adds three points and you change nothing

United States restaurant inflation already hit 8.8% in March 2023 (National Restaurant Association); whoever recosts quarterly absorbs the blow, whoever recosts once a year finds out from the balance sheet. Start Monday with recipe cards and finish the quarter with variance under 1.5 points: that is the whole plan. Days 1 through 30, weigh and document real yields on your twenty main inputs and build recipe cards with a waste factor; days 31 through 60, install biweekly physical counts and calculate your first inventory-based food cost, which almost always stings; days 61 through 90, close the variance plate by plate and recost the menu. Three indicators for the board: inventory-based actual food cost, theoretical-to-actual variance in points, and contribution margin per plate in dollars, which is what decides who stays on the menu. With 99% of operators reporting higher labor spend in 2024 (TouchBistro), food cost is the only large line still answering to internal discipline.

Chapter 8 — A 90-day roadmap with KPIs a board of directors can read

Call tomorrow's protein count. Theoretical cost comes from the recipe, never from the invoice. Buying a kilo of loin at 18 dollars does not mean the served kilo costs 18: after trimming, fat and cooking loss, net yield can land at 68%, pushing the served kilo to 26.47 dollars. Costing on the invoice instead of on yield is the arithmetic origin of most of the gap, and it is an error of arithmetic, not of judgment. Actual cost is measured by inventory difference, never by adding invoices. The formula is opening inventory plus period purchases minus closing inventory, divided by sales of THAT same period. Adding up the month's invoices ignores what stayed in the walk-in and what was consumed from the prior month, producing a number that swings month to month for no reason. Variance is the diagnosis, not the result. A restaurant can post 31% actual food cost and be bleeding, if its theoretical was 26%.

Chapter 9 — Five differences between measured food cost and invented food cost

Another can post 34% and be perfectly healthy, if its theoretical was 33.5% because it sells premium product at a high ticket. The absolute percentage without its theoretical beside it informs nothing. Payroll, rent and utilities do NOT load onto the plate. This is a hard rule of the Masterestaurant method and the source of endless arguments with accountants trained in manufacturing. A restaurant is not a factory: payroll is installed capacity, not a variable input per unit. Loading it into the plate produces an apparent 50% food cost that makes any sensible menu decision impossible. The 32% ceiling is a maximum, not an objective. At Masterestaurant we work with 28%-30% as the healthy operating band for full service and tolerate up to 32% in premium-product concepts with a high average ticket. Above 32%, the remaining contribution margin cannot cover Prime Cost without compressing EBITDA.