Food cost: the mistakes you pay for monthly vs the right method

Why food cost calculated on purchases lies by three to six points?

Because a month's purchases almost never match what the kitchen actually consumed that month, and the gap swallows between three and six margin points without triggering a single alert.

The honest formula is opening inventory plus purchases minus closing inventory, divided by food sales for the SAME period; anything else is an estimate dressed up as data. At the three-unit Bogotá group we reviewed in March 2026, with 412 million pesos in quarterly food sales, the accountant reported 29,4 % while real consumption came in at 34,8 %: five and a half points that, on that base, mean 22,6 million pesos evaporated in ninety days. With labor already eating 25 % to 35 % of revenue according to the Bureau of Labor Statistics, no operator has the cushion to finance an arithmetic error of that size. A healthy average food cost can live comfortably alongside two or three dishes that bleed, which makes the global number the worst possible place to hunt for the problem.

The weighted average lies: break food cost down by product family

Breaking the Bogotá case down by family surfaced the usual pattern: proteins at 41 % against a theoretical 33 %, an eight-point deviation, while beverages —typically running between 18 % and 22 %— and desserts pulled the average down and hid the hole for months. Profitability gets decided dish by dish, on the recipe card and on the scale, never in the spreadsheet cell that sums up the month. Rank your SKUs from largest to smallest deviation against theoretical and work only the top five: in most menus, those five explain more than half of the total overrun. The decision this block of numbers triggers is simple: ban any aggregate food cost report that arrives without its family-level breakdown. Theoretical comes from multiplying units sold by recipe-card cost; actual comes from consumption measured with inventories. The distance between them is NOT an accounting error: it is waste, ungrammed portions, theft, sloppy receiving and remade plates, all bundled into one figure.

The theoretical-versus-actual gap is the only honest thermometer of line discipline

A gap under two points signals a disciplined kitchen; two to four points means standardization work is pending; above four, the problem stopped being your supplier and became your internal process. Run it backwards for a second: if your actual food cost dropped two points tomorrow without touching a single purchasing negotiation, that entire gain would come from the line, and on annual sales of 1.600 million pesos it would be 32 million in extra profit without selling one more plate. Measure the gap weekly across your twenty highest-rotation references. A 30 % food cost can be excellent or suicidal depending on what payroll weighs beside it, because what decides survival is prime cost, the sum of food and labor. The National Restaurant Association measured a 2024 median payroll cost of 36,5 % of sales in full service, well above its historical averages, and Toast places the labor band between 25 % and 35 % depending on format.

Food cost is never judged alone: it lives inside prime cost

Add it up: food at 34,8 % plus payroll at 36,5 % puts prime cost near 71 %, and with occupancy inside the 6 % to 10 % of gross sales band Toast recommends, plus utilities at 2 % to 5 % of total revenue, operating profit disappears before it reaches the line. The uncomfortable, firm conclusion: cutting food cost without looking at payroll just moves money from one pocket to another. A dish running 32 % raw-material cost in the dining room can reach 50 % effective cost on delivery, and anyone who skips that recalculation loses money on every order they celebrate. OPA! quantified the true total cost of third-party delivery in 2026 —commissions, promotions, refunds and hidden fees included— at 30 % to 40 % of order value, so on a 60.000-peso ticket carrying 19.200 in raw material, the platform takes another 18.000 to 24.000. Under 20.000 pesos remain to pay kitchen, packaging, energy and rent.

Third-party delivery: the dish costs the same, your margin collapses anyway

I got this wrong for years, recommending mirror menus at identical prices; today my recommendation carries no qualifiers: build a dedicated delivery menu with low-food-cost references, grammage adjusted to the packaging, and prices marked up 15 % to 25 %. If your digital channel cannot carry that markup, that channel is not worth having. Food cost calculated on badly taken inventories cannot support any decision at all, and that is why so many management committees argue over figures nobody is able to defend. Manual counting at that three-unit group burned 14 labor hours a month and arrived with keying errors on 8 % of its lines; with computer vision over the scale and OCR reading of supplier invoices, counting fell to 3 hours and error dropped below 1 %. Eleven recovered hours a month matter little next to the other half: moving from 8 % to 1 % error turned a number people debated into a number people acted on.

Dirty data produces expensive decisions: capture stopped being optional in 2026

Diego F. Parra insists at Masterestaurant on that same running order, and he allows no shortcuts: fix capture first, measure the gap second, and only then touch a menu price. Take the scenario all the way out, because that is where the arithmetic shows. A restaurant at 33 % food cost absorbs a 12 % increase on proteins, which represent 45 % of its purchases; direct impact is 1,8 points and cost jumps to 34,8 %. On monthly sales of 400 million pesos, that is 7,2 million less profit each month, 86 million a year. Add energy moving —Toast calculates 2,90 dollars per square foot annually in electricity and 0,85 in natural gas— and the blow compounds. Now the part almost nobody does: recovering those 1,8 points does not require repricing the whole menu, only re-engineering the five references with highest volume and thinnest margin. Raise those five by 6 %, correct two grammages against the recipe card, and you will have recovered your point and a half without your regulars noticing anything.

The 3 numbers you should tattoo on yourself

Three numbers, each with its concrete action. FIRST: 34,8 % versus 29,4 %, the distance between consumption-based and purchase-based food cost in the Bogotá case — action, change your report formula today to opening inventory plus purchases minus closing inventory over same-period sales, and recalculate the last six months before making any pricing decision. SECOND: 36,5 % payroll on sales in full service, per the National Restaurant Association 2024 — action, add that number to your food cost every Monday and watch that prime cost never crosses 65 %. THIRD: 30 % to 40 % of the order goes to third-party delivery, per OPA! 2026 — action, build a separate digital menu with a 15 % to 25 % markup or shut the channel down. Start with the first one this week: without it, the other two rest on sand. The first difference is arithmetic and the most expensive of the lot: purchases and consumption are not the same thing.

The three differences that move margin

If you bought a three-month drum of olive oil in January, January's purchases inflate food cost and February's deflate it, and neither figure describes what the kitchen actually spent. Under the consumption method that oil hits cost as it gets used. In the groups we have worked with, fixing this alone drops monthly volatility from ±6 points to ±1.2, and only then does any pricing decision become possible. The second is the theoretical-actual gap, which almost nobody measures and which is the only honest thermometer of discipline on the line. Theoretical food cost comes from multiplying units sold by recipe-card cost; actual comes from consumption. When the gap clears two points there is waste, over-portioning, theft or stale pricing, and the order of investigation matters: expired recipe cards first, then portioning, then goods receiving, and only at the very end do you talk about theft.

The three differences that move margin — in practice

Starting with theft wrecks the team's trust and is almost never the main cause. The third is recosting frequency, and in 2026 that is where AI genuinely changed the game. Recosting 60 recipe cards by hand takes six to nine hours, so it happens once or twice a year; by the time you finish, three inputs have already moved. An engine that reads supplier invoices with OCR and rewrites each card's cost the moment it detects a swing above 5 % turns that project into a continuous process. The owner no longer chases the number: the alert arrives saying octopus is up 18 %, and the decision is whether to raise the price, swap the garnish or pull the dish.