How to calculate restaurant food cost: the expensive mistake and the alternatives that hold up

The one honest formula, and why the recipe sheet lies

Actual food cost = (opening inventory + purchases for the period − closing inventory) ÷ net food sales for that same period, and there is no valid second version of that math. A grill house in Bogotá sold 118 million pesos in July, certain of a 29% figure written into a 2024 recipe sheet; the physical closing count came back at 34,7%, nearly six points that on those sales meant almost seven million pesos gone without a single dish given away on purpose. Dividing a recipe's theoretical cost by its menu price answers a different question entirely: what the plate SHOULD cost you if everything ran perfectly. The average restaurant wastes between 4% and 10% of the inventory it buys (The Restaurant HQ, 2025), and that whole range lives inside the gap between the two calculations. Your business gets decided by the second one, the one that counts what got thrown out.

When the recipe sheet stops being enough (and the number that gives it away)?

Your costing sheet stopped serving you the day its gap against physical inventory passed two percentage points for two consecutive months, and almost nobody tracks that signal.

At the grill house, three leaks explained the spread: a pantry uncounted since February, an out-of-calibration line scale sending out breaded chicken with 30 extra grams per portion, and 41 cancelled delivery orders the kitchen actually produced that nobody deducted anywhere. No spreadsheet catches that, because a recipe sheet assumes zero waste, exact portions and no cancellations. Diego F. Parra insists at Masterestaurant on an order that sounds obvious and almost nobody honors: count first, cost second. I got this wrong for years, recommending flawless costing sheets with per-gram detail, until I accepted that without a physical count they are fiction with good typography. Counting the entire pantry on the last day of every month is the cheapest option available and the one no restaurant should skip, whatever it costs in hours.

Option 1: full monthly inventory, the absolute floor

It takes three to five hours of two people, a frozen list of SKUs and prices from each item's latest invoice; the switching cost is essentially zero in money and high in discipline, which is exactly where operators fail. It fits the owner of a single location with 30 to 45 menu SKUs and an orderly weekly purchase. Its limit: twelve snapshots a year never tell you WHICH week your margin walked out, so you find the hole thirty days after it opened. At that grill house, six points of variance meant 6,9 million pesos lost before anyone weighed a single kilo of beef. Anyone unable to stop operations weekly has a shortcut with evidence behind it: count only the 10% of SKUs carrying 60% to 70% of your food cost. Protein, cheese, oil, liquor and the star ingredient of your best-selling plate usually cover it. That brings the count down to 40 minutes on Monday before doors open, and gives you 52 readings a year instead of 12.

Option 2: weekly count of the ten SKUs that drive your cost

The profile that needs this: an operator with heavy delivery volume, where margin shifts week by week rather than month by month. Ground beef at 80-90% lean went from 4,56 to 5,63 dollars per pound between 2025 and mid-2026 according to the USDA, nearly 23% in twelve months; on a monthly count you absorb that hit thirty days before it shows up on paper. Always divide by NET settled sales, with commission already deducted, because using gross guest-facing sales hands you an artificially low food cost and you celebrate a margin that never reached the bank. At 25% commission, a 60.000-peso order lands in your account as 45.000; cost against the 60.000 and your percentage lands five or six points below reality. Commission is not raw material: it belongs in channel expense, its own P&L line, exactly like packaging. Every delivery carries between 1.200 and 3.500 pesos of container, lid, bag and tamper seal, and that cost scales with each order your geotargeted ads generate within three kilometers.

The delivery trap: the sales figure the platform reports is not your revenue

Burying it in food cost pollutes the metric; leaving it off the income statement makes it vanish, which is worse. A system that deducts recipes against sales in real time hands you daily theoretical food cost and automatic variance against your count, and runs between 80 and 250 dollars monthly per location plus 40 to 80 hours of initial setup. The heavy lifting is not the software: it is loading every recipe with real gram weights measured in your own kitchen, not the ones your chef recites from memory. It suits operators with two or more locations, or menus above 60 SKUs, where manual counting no longer holds. Its risk is dangerous precisely because it feels comfortable: load the recipes wrong and the system returns a beautiful, daily, false number, carrying all the authority a screen confers. Variance still demands the physical count. Software speeds up the reading; it never replaces the scale.

The grill house counterfactual: what a weekly count would have changed

Had that grill house counted its pantry every Monday since March, the variance would have surfaced in the second or third week, when the hole was worth 400.000 pesos rather than 6,9 million. Calibrating the scale runs about 150.000 pesos, checking the breaded chicken portion takes half an hour of one cook, and closing the cancellation leak only requires cross-checking the platform report against kitchen tickets. Five blind months turned three thirty-minute problems into a seven-figure crater. Here sits the paradox of the trade: the restaurant with the least time to count is exactly the one bleeding the most from not counting, and it escapes by counting fewer SKUs, not by finding more hours. Foodservice accounts for 17,9% of United States food surplus (ReFED, 2024), and full-service operations contribute more than 43% of that total. If your monthly food cost has moved less than a point for six months and your variance against theoretical stays below 1,5%, keep the monthly count and buy nothing.

When NOT to change your method?

Operators buy inventory software for a 28-SKU location with daily market purchasing, where the system adds two hours of weekly data entry and reveals nothing the owner did not already know.

Measurement sophistication should follow the size of the mess, never the other way around. Do not switch during peak season or through a menu change either: you need a stable period for the new reading to mean anything. And if you still lack updated per-item prices and a frozen SKU list, no tool will save you. This week, before buying anything, calibrate the line scale and weigh three portions of your best-selling plate. The revenue Rappi or Uber Eats reports is gross customer spend, not your income. Divide food cost by that inflated figure and your percentage lands artificially low, so you celebrate a margin that never reached the bank. Always divide by the NET amount settled to you, commission already deducted, and book the commission where it belongs: channel expense, not raw material.

Where the math breaks in a restaurant running on the digital local channel?

Delivery packaging is its own category. An average order carries between 1.200 and 3.500 pesos of container, lid, bag and tamper seal, and that cost scales with every order your geotargeted ads generate.

Push it into food cost and the indicator gets dirty; leave it out of the P&L and it vanishes. It gets its own line, next to commission. The promotions the platform algorithm demands —two for one, subsidized free delivery, a 30% discount to enter the carousel— are discounts against revenue, and they erode the denominator of your formula. One aggressive campaign month can lift food cost four points with no supplier change and no recipe change. Orders cancelled after cooking never register as waste in any point-of-sale system, yet the food did leave your storeroom. Only the physical count exposes them, and in high-volume delivery operations they account for 1% to 3% of tickets.

Where the math breaks in a restaurant running on the digital local channel — in practice?

Cost per review and local ad spend are not food cost, though plenty of owners blend them when asking what it costs to sell a plate.

Keep them apart: food cost governs the kitchen, acquisition cost governs marketing, and confusing them makes both unfixable.