How to calculate restaurant food cost: the numbers before and after measuring it properly

What is the correct real food cost formula?

Real food cost equals opening inventory plus period purchases minus closing inventory, divided by food sales for that same period, multiplied by one hundred.

No averaging invoice prices, no pulling it from the recipe: the storeroom decides. A 180-cover steakhouse in Bogotá believed it was running at 29% because the chef had estimated it on a spreadsheet using purchase prices from fourteen months earlier; once we closed a real 28-day inventory, the figure came in at 37,4%. Against monthly food sales of 214.000 USD, those eight points are 17.100 USD evaporating every thirty days with no owner and no explanation. The decision that comes out of this formula is not cosmetic: if your number clears 34%, the first move is not raising prices, it is counting the storeroom two periods in a row and comparing.

Plate cost: yield loss is always loaded in

Per plate, the calculation is the sum of each ingredient's cost WITH its yield loss loaded in, divided by the selling price before tax. Yield loss is the part almost nobody loads, and that is why almost nobody gets it right: a striploin with 18% trim loss does not cost what the invoice says, it costs 22% more per usable kilo. My rule on costing sheets is simple and a bit stubborn —load the yield loss measured in YOUR kitchen, not the supplier's and not the manual's—, because the spread between two kitchens working the same cut reaches six or seven points. A 24 USD plate with 7,20 USD of ingredients gives 30% on paper; the same plate with real yield loaded rises to 8,40 USD and lands at 35%. That five-point jump decides whether the menu holds the margin or drains it quietly.

Variance is the number that actually moves EBITDA

The gap between theoretical cost and real cost —food cost variance— is the only cost metric that correlates directly with EBITDA in a full-service operation. Not average check. Not table turns. The subtraction. Theoretical answers what the plate SHOULD cost; real answers what left the storeroom, and an owner who knows only the first is reading an intention rather than a result. Below 2 points of variance, the operation is under control. Between 2 and 5, there are orderly leaks that portion control and training fix. Above 5 there is theft, overportioning or off-standard purchasing, and no fourth explanation exists however hard you look for it. With full-service margins running between 3% and 8% per WhippleWood CPAs (Restaurant Financial Benchmarks 2026), five points of variance swallow the entire profit.

Every 28 days, even if the count comes out imperfect

Closing frequency matters more than closing precision, and this is the part where for years I recommended exactly the opposite. A count that is 92% accurate repeated every 28 days is worth far more than a flawless count taken once a year, because the correction arrives while portion size, price or supplier can still change. Thirteen annual closes give thirteen chances to adjust; two closes give two, and by the time the second one lands the whole season is gone. The paradox resolves itself the moment you accept that inventory is not an accounting exercise but an alarm system: an alarm that rings late with laboratory precision is useless. Diego F. Parra and the Masterestaurant team have taken this measurement into operations across 43 countries, and the pattern repeats so often it is nearly boring: whoever does not close every 28 days does not know their food cost, they guess it.

2026 inflation is moving the floor under your feet

A flat food cost in 2026 means you improved, because the floor is shifting upward. The USDA Economic Research Service, in its June 2026 Food Price Outlook, projects a 3,2% rise across all food, with 2,8% at the supermarket and 3,6% for food away from home, against a historical average of 3,5% per year. Averages lie by category, though: beef climbs 7,5% at retail and 9,4% at wholesale with the cattle herd at a 75-year low, while coffee and non-alcoholic beverages advance 5,7%. A steakhouse with 40% of its purchasing in red protein is not facing 3,2%, it is facing close to 5 weighted points. The concrete decision: if your menu leans on beef, reprice plate by plate each quarter instead of once a year, or shift volume toward lower-demand cuts before the margin narrows.

How to read these numbers in YOUR operation?

Benchmarks read differently by size, and applying them wholesale is the costliest mistake around.

In a small single-unit restaurant, with full-service margins of 3% to 8% (WhippleWood CPAs, 2026), the variance target is 2 points and the owner counts in person every 28 days: no structure exists there to absorb a 5-point leak. In a mid-sized operation of two to five units, consolidated food cost deceives because it averages a healthy site with a sick one, so measure per unit and compare units against each other before comparing against the sector. In a group of six or more, the relevant number stops being food cost and becomes the DISPERSION between units: if your best site sits at 29% and the worst at 36%, you have a standards problem rather than a cost problem, and recipes and scales fix it, not purchasing negotiations.

Where these benchmarks come from and where they stop working?

It is worth saying where these figures come from and where they stop being useful. Inflation data comes from the USDA Economic Research Service (Food Price Outlook, June 2026), which measures the United States market:

the direction is useful anywhere, the magnitude does not transfer intact to Mexico or Colombia, where exchange rates and logistics weigh more than agricultural supply. The margin ranges come from WhippleWood CPAs (Restaurant Financial Benchmarks 2026) and bundle thousands of very different operations under three coarse labels. And the Mexican market, with over 641.000 restaurants contributing 1% of GDP according to CANIRAC and INEGI (2024), carries a fragmentation no average benchmark captures well. Treat them as a thermometer for direction, never as a target: your target comes from your own four-period history, not from somebody else's table.

What happens if you do not measure: the full scenario?

Suppose you decide not to count the storeroom for a full year. The first quarter shows nothing visible, because cash comes in and bills get paid;

by the second, the 3,6% inflation in food away from home (USDA ERS, 2026) starts eating the margin without surfacing in any report, and overportioning becomes normal since nobody corrects it. By the third, variance already sits at 6 or 8 points and the owner blames sales. In the fourth it gets fixed with layoffs, which is the signal that you arrived late. All of that is avoided with a two-hour count every 28 days, and investing in waste prevention returns 7 USD for every dollar put in according to ReFED. Count your storeroom this Friday, run the formula against food sales for that same period and write the number down. That figure, repeated thirteen times a year, beats any consultancy.

The four differences that move cash

Theoretical cost answers «what SHOULD this have cost»; actual cost answers «what left the storeroom». An owner who only knows the first is reading an intention rather than a result, which is why variance —the subtraction between them— is the number worth chasing. Under 2 points, the operation is controlled; above 5, you have theft, over-portioning or off-standard buying, and there is no third option. Frequency beats precision. A count that is 92% accurate done every 28 days is worth far more than a perfect count done once a year, because the correction arrives while it can still change something. Thirteen closes give you thirteen chances to adjust price or portion; two closes give you two, and by then the season is gone.

The four differences that move cash — in practice

Food cost never travels alone: read it glued to the dish contribution margin. A dish at 34% carrying 21 USD of absolute margin pays payroll better than one at 24% carrying 6 USD, and the mistake I see over and over is cutting the «expensive» dish that was holding up the break-even point. Menu engineering happens with both columns visible, never with the percentage alone. Payroll, rent and utilities do NOT get loaded onto the dish. They belong to the break-even point, where you decide whether the month's volume covers the structure. Mixing them produces that inflated 48% food cost some consultants present as a finding, when it is a method error whose side effect is an owner raising prices where he should not have.