Merma por robo hormiga: the before and after of auditing your theoretical vs actual food cost

Step 1: split the hand that buys from the hand that counts

Start by breaking apart a function that almost always lives in one person: whoever buys cannot be whoever counts inventory, because the system then loses its only contrast and variance gets dressed up with no effort at all. The deliverable here is a one-page memo with two names, two signatures and a weekly cutoff date, and you verify it by checking that count sheets carry the signature of someone who appears on no purchase order from the same period. It costs nothing and usually hands back the first point of food cost. In a restaurant billing 60,000 USD a month, that point is 600 USD monthly, 7,200 a year, money already sitting in the register that walked out without an invoice. If your operation is too small for two hands, you do the counting. Frequency beats precision, so the second step shrinks the universe: rank every reference by monthly consumed value, flag the ones carrying 80 % of spend —usually between 15 and 25 lines— and leave the rest out of the weekly count.

Step 2: pin down your 20 class-A items and ignore the other 280 for now

An imperfect weekly inventory over 20 items catches a leak in seven days; a perfect monthly one over 300 references catches it after two months of profit are gone. The deliverable is a signed list with those 20 lines, their counting unit and current unit cost, verified by adding up: that list's monthly value should land between 75 % and 85 % of your purchases. Beef deserves particular attention this year, since the USDA projects wholesale beef prices climbing 9.4 % during 2026. BLIND inventory shifts the result more than anyone expects, because when the count sheet arrives with the theoretical balance printed on it, the eye confirms the expected figure and counting becomes paperwork. Print the sheet with item, unit and an empty box, nothing else. Over the first three weeks you will see gaps of 8 to 15 % on your highest-value items, and those gaps are not a failure of the method but the first time you are looking at reality.

Step 3: count blind, with no system balance in sight

The deliverable is a signed count sheet, with start and finish times, loaded into the system that same day; verify it by confirming no box arrives pre-filled with the system's round number. I got this wrong for years: I believed handing over the balance saved time. Now you read the number, and you read it as money. Take the theoretical cost your recipe book throws for the week's sales, subtract the actual cost your inventory throws, and that gap is your variance. A 60,000 USD monthly restaurant with theoretical food cost at 29.4 % and actual at 33.1 % —the Bogotá case I opened with— loses 3.7 points, meaning 2,220 USD a month and 26,640 a year, more than an executive chef's salary. Percentages anesthetize; dollars force a decision. The deliverable is a weekly four-column table: item, theoretical cost, actual cost, difference in USD, sorted from largest loss down.

Step 4: calculate variance in dollars, never in percentage

You verify it when that column's total matches the hole in your P&L. Shrinkage is NOT theft, and confusing the two sinks these programs. Shrinkage is every loss between purchase and sale: protein trim, spoiled product, returned plates, over-portioning, unrecorded comps and, yes, sustraction too. Petty theft is only a fraction of it. Before suspecting a person, assign each variance line to a measurable cause: weigh trim for five days, log comps in a notebook, measure each cut's real yield against its spec sheet. Over-portioning and trim usually explain more than half. The Masterestaurant financial pillar treats this as a measurement problem rather than one of trust, and Diego F. Parra insists on the order: first the measuring stick, then the number, and only at the end the people. Flip that order and the program dies in month two. First comes the manhunt: staff meeting, prosecutor's tone, new cameras.

The four mistakes that derail this guide

It wrecks morale, drives off your good cook and leaves the other five sources of shrinkage untouched. Second is blaming inflation, which moves purchase cost and shows up on the invoice —the USDA reported food prices rising just 2.3 % through 2024 and forecasts 3.6 % for food away from home in 2026, figures that cannot explain three points of food cost. Third is rewriting the recipe book mid-measurement, because variance then blends two causes and stops saying anything. And the fourth, costliest of all, is measuring one week, finding 2,200 USD of leakage and never measuring again; without weekly cadence the leak returns to its prior level within roughly forty days. Check seven things, and none of them accepts an "almost". Two different names sign purchasing and counting. The 20-item class-A list exists and covers between 75 % and 85 % of purchases. Your last four count sheets went out blind, with start and finish times.

Closing: how to know it all landed

The USD variance table has four consecutive weeks loaded. Every line above 100 USD carries a written cause beside it, not a person's name. Total prime cost holds under 60 % of sales, the healthy target per Toast and Restaurant365, with limited-service median sitting at 65 cents per sales dollar according to the National Restaurant Association. And week four's variance runs lower than week one's. Meet all seven and the leak stopped being a suspicion and became a figure you govern. Who counts inventory decides whether the data is worth anything. When the buyer also counts, the system loses its contrast and variance can be dressed up effortlessly; separating those two hands costs nothing and usually explains the first recovered food cost point. BLIND counting —counting without seeing the system's theoretical balance— changes the result more than anyone expects. Counting in sight of the balance, the eye tends to confirm the expected figure; counting blind, differences of 8 to 15 % surface on the highest-value items during the first three weeks.

Five differences that decide whether the number drops

Frequency beats precision. An imperfect weekly count on 20 class-A items catches a leak in seven days; a perfect monthly count on 300 SKUs catches it after 2,000 USD are gone. Choose speed over completeness, every time. Your recipe book is the measuring stick, and a miscalibrated stick ruins the whole measurement. If theoretical cost runs on old prices or ignores yield factor —the tenderloin arriving at 1.2 kg and yielding 0.86 kg usable—, much of the gap you are chasing is an accounting error rather than a leak. Closing the bar is a separate job from closing the kitchen, with its own unit of measure and its own rhythm. A bar without ounce control runs 15 to 25 % variance quite normally, and that single front can outweigh all five kitchen shrinkage sources combined wherever liquor passes 30 % of sales.