Plate costing: the before and after of measuring actual cost against theoretical

Chapter 1 — The gap between theoretical and actual cost is your real P&L

Food cost variance —the difference between what your recipe says the plate cost and what the supplier invoice says it actually cost— is the only costing number that decides whether you make money this month. A menu costed at 30% that buys at 34% is giving away four points of sales, and in an operation doing 800 thousand USD a year that means 32 thousand USD evaporating with no line of its own in the P&L, buried inside «cost of goods sold». The National Restaurant Association reported that 90% of full-service operators raised prices during 2024 and 60% pulled dishes off the menu, two levers applied blind when nobody reconciles theoretical against actual. I worked with the annual spreadsheet for years and I was wrong: the pretty number on the recipe card kept me calm while the register said something else. Weekly reconciliation, plate by plate, is what turns costing into information.

Chapter 2 — Why doesn't food cost percentage tell you which plate to push?

Percentage watches risk; contribution margin in dollars decides the menu. Two plates with an identical 30% food cost but 9 and 22 USD of unit contribution are not comparable:

if your server sells the first one, that table leaves 9 dollars to cover payroll that already runs above 25% of expenses according to Toast (2024) and that in profitable full-service operators sits at 34,2% of sales against the 36,5% average, per the National Restaurant Association Restaurant Operations Data Abstract with 2024 data. Those 2,3 points of difference are not won by negotiating wages, they are won by selling the right plate. The practical consequence: rank your menu by absolute contribution, not by percentage, and keep the percentage for triggering purchase alerts when an input moves. Costing 70% of the menu is not costing the menu, it is estimating it with a serious name.

Chapter 3 — Incomplete coverage: the recipe card that leaves out whatever wastes most

What almost always gets left out —sides, mother sauces, in-house bakery, comps— is precisely what absorbs waste without traceability, and in full service the weight is real: ReFED (2024) attributes more than 43% of total foodservice food surplus to full-service restaurants. A sauce that yields 40 portions in theory and 33 in practice makes every plate that uses it 21% more expensive, and that overcost travels invisibly across ten menu items. The rule inside the Masterestaurant framework is simple and hard: no intermediate preparation enters production without a card whose yield was measured in your own kitchen, with your trim loss and your cooking loss, never the textbook yield. Continuous costing is video and annual costing is an old postcard. Brazil holds close to 38% of world coffee supply according to Bellwether Coffee, so a frost in Minas Gerais moves your hot beverage cost weeks before you open the spreadsheet in January.

Chapter 4 — Frequency: an annual cost is a photograph of a market that moves every week

The same happens with protein, oil and dairy. The architecture Diego F. Parra recommends at Masterestaurant wires together three sources almost nobody crosses: the recipe catalogue, the supplier purchase history and POS sales, then lets an AI engine recalculate theoretical cost every time an invoice lands, flagging only the plates whose cost moved more than 3 points. You don't review 120 cards, you review the six that changed. That is the difference between a system and an archive. Size changes the lever, not the formula. Below 500 thousand USD the owner cooks and buys, so costing lives in the recipe card and in four critical inputs: a well-built spreadsheet is enough, and chasing one point of food cost equals 5 thousand USD a year. Between 500 thousand and 1 million the classic 3 to 6 point gap between theoretical and actual shows up, and weekly inventory stops being optional.

Chapter 5 — The same costing weighs differently in each revenue band

Above 1 million, with two or three locations, the problem is variance BETWEEN units, not the average. Over 5 million costing becomes purchasing governance with contracts and tiered pricing, and above 10 million it turns into a valuation matter: the sale multiples Sofer Advisors reports, between 2,80x and 3,65x EBITDA, punish every uncontrolled point of food cost directly. Above 5 million USD, costing in fine dining and large-format themed venues breaks at the top, not at the bottom. A tasting menu with 30% of items rotating each season turns every menu change into a full costing project, and R&D waste —the 14 versions of the dish that never sold— fits on no recipe card. Add specialized labor in an industry where 99% of operators reported spending more on payroll during 2024 according to TouchBistro. The market punishes it at exit: Sofer Advisors places fine dining multiples at 2x to 4x EBITDA against 4x to 7x for fast-casual.

Chapter 6 — High end: the celebrity-chef restaurant pays costs the standard spreadsheet never sees

The reading I defend is uncomfortable: fine dining must cost the COMPLETE MENU as a unit, with creative waste prorated, rather than plate by plate. With 20% input inflation and a 30% theoretical food cost, your actual cost jumps to 36% and you lose six points of margin on sales. Translated: a restaurant doing 1 million USD a year that ran on the 10,66% pre-tax margin NYU Stern reported as the sector average in 2024 drops to 4,7%, and if your real margin was the 12%–13% after tax that WhippleWood CPAs records for publicly traded chains, you keep half. At 5% inflation the loss is 1,5 points and gets absorbed; at 12% it is 3,6 points and the menu needs re-engineering; at 20% no price adjustment covers it without losing traffic. The stress scenario is not an academic exercise, it is the reason costing gets reviewed every week.

Chapter 7 — A 90-day rollout with KPIs a board actually understands

The plan that works puts reconciliation before technology. Days 1 to 30: recipe cards for 100% of the menu with yields measured in your kitchen and not in the book, intermediate preparations included. Days 31 to 60: weekly inventory of the 20 SKUs that carry 80% of spend and a first variance measurement, which in most full-service operations lands between 3 and 6 points. Days 61 to 90: automated invoice capture and per-plate alerts. Three KPIs for the board, only three: theoretical-actual variance in points, average contribution per guest in dollars, and recipe card coverage as a percentage. The multiplier ABRASEL calculated for Brazil —every R$1.000 spent in bars and restaurants injects R$3.650 into the economy— explains why this trade sustains so much employment and why every point you recover matters beyond your own register. Start Monday with the ten recipes you sell most.

Chapter 8 — Six differences that move the margin

FREQUENCY. An annual cost is a photograph of a market that moves weekly; continuous costing is video. With Brazil supplying roughly 38% of the world's coffee (Bellwether Coffee), a frost in Minas Gerais shifts your hot beverage cost long before you reopen the spreadsheet. UNIT OF MEASURE. Food cost percentage watches risk; contribution margin in dollars drives decisions. Two dishes at 30% food cost yielding 9 and 22 USD are not the same dish, and the menu should push the second one hard. COVERAGE. Costing 70% of the menu leaves out precisely what wastes most: garnishes, sauces and in-house bakery. An incomplete spec sheet turns costing into an estimate wearing a serious name. SPEED OF RESPONSE. Catching a deviation in seven days lets you fix portion, supplier or price inside the same month; catching it in forty-five days only lets you mourn it. DATA GOVERNANCE. With 98% of operators reporting higher labor costs (National Restaurant Association) and payroll above 25% of expenses (Toast, via Restaurant Dive, 2024), prime cost needs an auditable historical series rather than a remembered average.

Chapter 9 — Six differences that move the margin — in practice

USE OF AI. Artificial intelligence does not price your menu: it kills data entry, normalizes units across suppliers, spots the input that moved and hands you back the time for the one thing machines still cannot do, which is menu engineering with judgment.