Restaurant inventory control: the weekly-count myth and the four alternatives that actually hold your margin

When full weekly counts stop paying for themselves?

Full weekly inventory stops paying for itself the moment your gap between theoretical and actual food cost holds still for three months straight, because a frozen gap is no longer information, it is a toll you keep paying in payroll.

The evidence is easy to gather: write down each week's variance and, if the last twelve swing less than one percentage point among themselves, the count stopped teaching you anything and merely confirms a flawed method. At that 92-seat Bogotá grill, the gap sat between 6.4 and 6.9 points for six months, an hour and a half of two people every Sunday, roughly twelve payroll hours a month buying a figure nobody ever used to change a purchase order, a portion size or a menu price. Counting only the four or five families that carry the spend — protein, cheese, liquor, oil — captures 70 to 80 % of your cost deviation with less than half the effort of a full count.

Option 1: daily spot count of critical families

Its profile: 40 to 120-seat operations where the owner or the chef still works the line and nobody can hand twelve monthly hours to a clipboard. Switching costs little, twenty to thirty minutes at close plus one afternoon to define what belongs in each family, and the payback shows up fast because the signal arrives the same day, while the shrink can still be corrected. It carries an honest limit: you will see the ribeye walked away, you will not know whether it was portioning, a short delivery or a generous knife, and that question belongs to the recipe card, not the spot count. Costing every recipe to the gram, trim loss included, attacks the cause while you can still fix it, which is worth more than any decimal measured after the fact. It suits kitchens with a stable 25 to 45-dish menu and purchases concentrated in a few suppliers; it does little for a kitchen rewriting half its menu monthly.

Option 2: recipe cards costed to the gram

Upfront effort is real: budget twenty-five to forty hours to build forty cards, plus a quarterly pass whenever inputs move, and that pass is not optional in a market where ground beef hit $5.63 a pound against $4.56 the prior year (USDA, 2026 price data) and arabica touched $4.41 a pound in February 2025 (Bellwether Coffee). Without living cards, every theoretical calculation ages within weeks. Perpetual inventory that depletes ingredients each time a dish rings up produces an elegant, false theoretical number whenever recipe cards are approximate, and that is its main risk, because owners believe it precisely for coming out of software. It works in limited-service chains with sealed recipes and industrial portioning; in Canadá that segment holds 46.4 % of foodservice sales against 43.1 % for full service (Statistics Canadá, 2024), and it is no accident the tool was born there. The module runs $60 to $200 per location per month, plus the labour of keeping cards current.

Option 3: perpetual inventory inside the POS

Diego F. Parra insists on an order that Masterestaurant does not negotiate: costed card first, module second; reversed, you automate an error and multiply it by every ticket. A full count once a month, closed against the period's invoices with no receiving that day, remains the only way to catch systematic storeroom theft, which none of the other three options can see. The reason is structural: the daily spot check watches families, the recipe card watches formulas and the POS watches sales, yet nobody physically compares what came in against what remains. Consider what happens if a supplier bills twelve cases and drops ten for eight months running; the theoretical figure never flinches, weekly variance absorbs it as noise, and by the time somebody finally counts, accumulated loss exceeds what a whole year of counting cost. For a 90-seat room, three or four hours a month and one well-built sheet close that door.

What measuring actually costs, in hours and money?

Before choosing, price the measurement itself: a full weekly count burns 8 to 14 payroll hours a month and that cost is fixed, whether or not the number gets used.

Weigh it against what is at stake. The average restaurant wastes 4 % to 10 % of the food it buys (The Restaurant HQ, 2025), and full-service establishments contribute more than 43 % of total foodservice surplus (ReFED, 2024), a sector that alone accounts for 17.9 % of U.S. food surplus. On a $30,000 monthly purchase, trimming three points of waste frees $900, more than those fourteen hours cost in almost any city in the region. Measurement justifies itself; what never justifies itself is measuring and then correcting nothing. Decide with four questions, in this order. First: who counts, and will that person still be here in twelve months? If the answer is you and nobody else, drop the full weekly count and keep the family spot check.

Four questions that settle which one is yours

Second: are your recipes costed to the gram, trim included? If not, the POS module will lie to you in beautiful typography. Third: does your menu turn over more than 20 % a quarter? If it does, invest in a simple procedure before investing in cards that will expire. Fourth: do you suspect physical shortages in the storeroom? If you do, the monthly full count is mandatory, however much it stings. The trade resolves its paradox here: the most precise method is not the best one, the best is the one your team repeats identically in week fourteen, when nobody is watching. Do not switch systems if your weekly count has run over a year on the same procedure and your theoretical-versus-actual gap sits under two points, because that habit cost more to build than any saving an alternative can promise you. Nor should you touch anything during peak season or an opening: changing your measurement method while the operation runs flat out leaves you without a historical series exactly when you need it, and you will spend three months rebuilding comparables.

When staying put is the right call?

And if your gap traces back to input inflation rather than sloppiness — remember food prices rose 2.3 % in 2024 (USDA ERS) and farm-level eggs 43.1 % that same year — the problem lives in your menu, not your count.

Raise the price, reformulate the dish, leave inventory alone. Full inventory measures the PAST accurately and blames nobody; recipe-and-check alternatives point at the cause while you can still fix it, which in a kitchen beats the decimal. The full weekly count charges its price in payroll — 8 to 14 hours a month — and that cost is fixed, while critical-families counting spends under half of it and captures 70 to 80 % of the cost deviation. POS perpetual inventory only works when recipe cards are costed to the gram; with approximate recipes it yields an elegant, false theoretical number, and owners trust it precisely because software produced it.

Where they genuinely part ways?

The alternatives fail against systematic storeroom theft, since without a physical count there is no way to see it; full inventory catches it, even if it takes two weeks.

Target food cost doesn't move with the method: 28-32 % per dish is the ceiling, never the goal, and no control system shifts that number when the recipe card doesn't exist. Prime cost — food plus total labor, with 60-65 % as the survival line — sits outside every count: it's governed by menu engineering and by the schedule grid. Par level is the only alternative that also cuts capital tied up in the storeroom, and in a venue holding USD 18,000 of average inventory that frees real cash.