How to calculate restaurant food cost: the recipe myth against the reality of your inventory

Which method wins when the two numbers disagree?

Actual cost measured through inventory counts wins, and theoretical cost stays as a pricing tool. The actual formula is short: opening inventory plus purchases for the period, minus closing inventory, divided by food sales for that same period.

If your storeroom opened at $12,000, bought $31,000 and closed at $11,500, you consumed $31,500, and against $90,000 in food sales that lands at 35% actual cost. Theoretical runs from the other end: multiply every plate sold by its recipe cost and add it up. That same month it may read 29%. The six-point gap is $5,400 that walked out of the storeroom without ever crossing a ticket. Actual cost is an accounting fact; theoretical is a promise. When the two disagree, the promise loses. Actual cost captures EVERYTHING that disappeared, whatever the reason; theoretical captures only what you wrote into the recipe. That asymmetry sets the division of labor.

What each method captures and what it cannot see?

A physical storeroom count absorbs cooking yield loss, 240-gram portions where the spec card calls for 180, unlogged comps, sloppy delivery packing and pilferage, all inside one number, which tells you how much you lost but not where.

The spec card, by contrast, is surgical plate by plate: it tells you your loin costs $8.40 and that at a $26 menu price it runs at 32%. Use it to price, to design the menu, to negotiate. What it cannot do is notice that the line cook is plating 60 extra grams, because a recipe has no idea what happened last night on the pass. Variance is the only number that tells you WHERE the problem lives, and it comes from subtracting theoretical from actual. Be stubborn about this arithmetic. A restaurant at 34% actual and 33% theoretical carries one point of variance: the operation is clean and the trouble sits in menu prices or purchasing, so the conversation belongs with the supplier and the menu.

Variance: the subtraction worth more than either figure alone

Another restaurant at that same 34% actual but 27% theoretical carries seven points: its spec cards are costed properly, its menu is priced properly, and seven points still leak on the line. Two opposite diagnoses behind an identical headline food cost, which is why 34% on its own means nothing. At Masterestaurant we always measure both, because the average alone cannot separate a purchasing problem from an operational hemorrhage. Actual costs you weekly discipline, theoretical costs you one heavy build, so start with actual. A storeroom count in a $90,000-per-month location takes 90 to 120 minutes when the shelves are organized by family, two people handle it Sunday after close, and it yields a usable number Monday morning. Costing 60 spec cards with real yields and trim loss, meanwhile, takes 25 to 40 chef-hours with a scale, and once built they need revisiting every time a supplier moves a price.

Setup cost: counting hours against desk hours

With the producer price index for all food sitting 35% above its February 2020 level (USDA ERS / BLS 2026) and headline PPI up 3.0% in 2025 (U.S. BLS), that revision is not annual: it is quarterly. Actual gives you a signal within a week; theoretical demands a month of work before the first data point. One $90,000-a-month restaurant closed the quarter at 35.2% actual against 28.3% theoretical, nearly seven points of variance, or $6,210 a month evaporating nameless inside cost of sales. The owner had spent three months negotiating proteins, sure the supplier was the culprit. We weighed twenty plates at random across two services: pasta was leaving with 118 grams of protein against the 90 on the card, ceviche with 145 against 120. Those two items alone, across 1,400 monthly covers, accounted for 3.8 points. Scales on the line, pre-portioned fish and a weekly count of the ten most expensive SKUs pulled variance down to 2.1 points in eleven weeks, and actual cost to 30.4%.

A pasta and ceviche house: $9,900 a year sitting in the subtraction

The supplier negotiation, which did happen afterward, added 1.3 points more. Variance first, purchase price second. Partial weekly counts beat the full monthly inventory, even though month-end is what your accountant wants. The reason is latency. A complete 400-SKU inventory reaches you up to 30 days late, and by then the money is gone; a weekly count of the 20 items holding 70-80% of storeroom value — protein, shellfish, aged cheese, liquor — takes 25 minutes and gives you an actionable pulse on Monday. I keep the full count for the last day of the month, because it reconciles with accounting and calibrates the weekly. I got this wrong for years: I counted everything, every 30 days, in a 400-line spreadsheet nobody read. That report was impeccably precise and nearly useless, because an exact number that arrives late changes no purchasing decision. A 24% food cost can be a worse business than a 33% one, and this trap repeats more than any other.

The paradox of a low food cost that wrecks the till

The percentage is a ratio, and you do not bank ratios: you bank contribution margin in dollars. A $14 salad costing $3.40 runs at 24% and leaves $10.60 of margin; a $34 lamb dish costing $11.20 runs at 33% and leaves $22.80. Shift your mix toward the salad because somebody decided to protect the percentage, and the month's food cost improves two points while profit falls, with the same covers and the same payroll hours. That is why the 32%-per-plate ceiling is a maximum, never a target, and why no menu decision gets made by reading the ratio alone. Measure the percentage to control leakage; decide the menu on absolute margin and turnover. If you can only sustain one method, sustain actual; if you can sustain a second, build spec cards for your 15 best sellers and stop there. For an independent house running one or two services, with fewer than 60 menu items, a weekly count of the 20 expensive SKUs plus a full monthly inventory covers 80% of the control in two hours of work.

What to choose for your operating profile

For a group of three or more locations, or any operation where delivery exceeds 25% of sales — remember that more than 40% of adults order delivery or takeout three to five times a month, per UpMenu 2024 — theoretical stops being optional, because without spec cards you cannot compare variance across units or isolate packaging overcost. Start this Sunday: count your ten most expensive items, save the number, repeat the following Sunday. Those two data points already give you a variance. The difference between these two methods is not academic, it is cash. A restaurant billing 90,000 USD monthly with 5 points of variance loses 4,500 USD a month, 54,000 a year, and it never appears on any line of the income statement under its own name: it hides inside cost of goods sold, as if it were a natural part of the business. I got this wrong for years, and I'll say it plainly: I believed a high food cost got fixed by negotiating with suppliers.

Variance: where the money you cannot see actually lives

Negotiating helps, of course, and a solid protein negotiation moves 2 to 4 points of that line item; but if your variance is 6 points, the supplier is not your problem and you will burn three months on the wrong conversation while the kitchen keeps plating 240 grams where the card asks for 180. The trade's paradox is that the two numbers most likely to contradict each other must live together. An owner who watches only theoretical lives inside an imaginary, profitable restaurant; one who watches only actual knows money is leaking but not from where, and ends up cutting blind —pulling the wrong dish, the one with worse food cost but better absolute contribution margin—. The bridge is measuring both, reading the subtraction, always over the same period with the same cutoff rule. On the ceiling, the house is strict: 32% food cost per plate is the MAXIMUM tolerable, not the target.

Variance: where the money you cannot see actually lives — in practice

Payroll, rent and utilities never load onto the plate —those belong to the break-even point of the whole business—, because once everything melts into a single figure you can no longer tell whether the problem is the kitchen, the lease, or CapEx from a remodel amortizing badly.