How to make a restaurant profitable: the costing errors that drain EBITDA and the method that repairs it

Chapter 1 — Prime cost is the only number that decides whether the business earns

A restaurant makes money when prime cost —food plus loaded labor— holds between 58% and 63% of sales, and not one point earlier. Everything else on the P&L follows from that. The 2026 arithmetic squeezes from both sides: USDA ERS forecasts food away from home at +3.6% and wholesale beef at +9.4%, with the U.S. cattle herd at a 75-year low, while the National Restaurant Association (2025) measured full-service wages and benefits at 36.5% of sales during 2024, well above the 33% that was the norm for two decades. Add 30% food to that 36.5% and you get 66.5%: three and a half points out of range before a single dollar of rent gets paid. That is the real starting point for almost any operation billing well and earning little today, and it explains why WhippleWood CPAs puts full-service margin between 3% and 8%.

Chapter 2 — Food cost variance: the money the recipe promises and the register never delivers

Between a dish's theoretical cost and its actual cost sits a gap almost nobody measures, and that is where the lost money lives. The recipe says the loin costs 6.40 dollars; the register records 7.90 because the cut yielded less than expected, the cook portioned by eye during the rush, two plates came back to the kitchen and someone walked product out the back door. That difference is food cost variance, and in operations without portion control it runs between three and six points of sales. Translated: a 900-thousand-dollar-a-year restaurant with four points of variance gives away 36 thousand dollars annually that it already charged the customer. The fix is cheap next to what it returns, since ReFED calculates seven dollars of benefit for every dollar invested in waste prevention. Scale on the line, standardized recipe, weekly count of the twelve items carrying 80% of spend.

Chapter 3 — The server-hour keeps yielding where cutting already hit bottom

Cutting has a floor; selling better does not, and that asymmetry decides where the effort goes after the first quarter. You can push food cost from 34% down to 31% by tightening purchasing and waste, but past that you start serving a worse product and the guest notices before your accountant does. Server-hour productivity, by contrast, allows sustained improvement: same dining room, same menu, same paid hours. The BLS reports median server wages of 16.23 dollars per hour including tips as of May 2024, and with California's minimum at 16.50 for tipped staff (Paychex 2025), every floor hour costs the same whether it sells or not. A server moving the average check from 32 to 36 dollars across 18 tables per shift generates 72 dollars of incremental revenue at ZERO labor cost. The difference is not charisma: it is knowing what to recommend and when to stay quiet.

Chapter 4 — Menu engineering: why a flat price increase destroys margin

Raising the whole menu 8% across the board is the single decision that has destroyed the most margin in the menus I review. It prices up the star dish —the one already converting and carrying the check— to the point where the guest starts comparing, and it leaves the dog untouched, sitting on the menu, eating mise en place, ordered by nobody. Menu engineering ranks dishes by popularity and dollar contribution margin, not by food cost percentage, and four different decisions come out of it: redesign, reposition, raise price, or cut. With coffee and non-alcoholic beverages forecast at +5.7% for 2026 per USDA ERS, beverage is the first category to re-engineer, because it tolerates price and its contribution margin exceeds 75%. A dish at 62% food cost with 14 dollars of contribution is worth more than one at 24% with 4.50. The register collects dollars, not percentages.

Chapter 5 — Revenue band completely changes which lever is the right one

The same diagnosis produces four different plans depending on what the house bills, and confusing them is the fast track to spending money where it does not return. Below 500 thousand dollars a year the owner IS the control system: standardized recipes and weekly counts rule there, no software, because segment margin sits between 3% and 8% (WhippleWood CPAs) and will not carry subscriptions. From 500 thousand to one million the first middle manager appears, and with him the portioning leak on shifts the owner does not cover. Above one million the lever shifts to sales forecasting by daypart, because with labor at 36.5% (NRA 2025) every badly scheduled hour costs 16 to 20 dollars. Above five million the problem stops being the plate and becomes rent and the lease: commercial rent in Los Angeles averaged 53 dollars per square foot per year in 2025 (Pepperlot), and a bad ten-year lease weighs more than any kitchen variance.

Chapter 6 — The high end above ten million pays costs the rest never meet

In the celebrity-chef house or the large-format themed concept, prime cost stops being the main enemy and three line items show up that simply do not exist below. First is the cost of the brand: royalties, chef fees, name licensing, taking 2% to 5% of gross sales before a single kilo gets purchased. Second is depreciation on an eight-to-fifteen-million-dollar build that must amortize while the novelty lasts. Third is the corporate structure holding the concept together. And all of it rests on volume, which is exactly what fails first: FAT Brands filed Chapter 11 in January 2025 with 2,200 restaurants open or under construction under its protection, according to Restaurant Business. Scale does not protect margin; it leverages it in both directions. Theoretical cost discipline per dish matters MORE here, not less, because each point of variance multiplies against volume that does not exist below.

Chapter 7 — The 90 days: what gets touched first and what waits

Sequence matters as much as the measures themselves, and the mistake that repeats most is starting with price. First thirty days: theoretical cost on the twenty dishes making 80% of sales, scale on the hot line, weekly inventory count and calculation of real variance. Days 31 to 60: menu engineering matrix with dollar contribution, menu redesign, and floor training on the six dishes worth recommending. Days 61 to 90: daypart scheduling against forecast and selective price adjustment in the categories that tolerate it, beverage first. Cornell measured that roughly 26% of new restaurants close or change hands in year one and around 60% within three, and the cause is rarely the food. This is the framework Masterestaurant applies when it walks into an operation that bills well and earns little, and Diego F. Parra reduces it to one rule: measure before you cut, because the problem almost never was the purchase price.

Chapter 8 — What happens if the gap stays open: the stress simulation

Take a 1.2-million-dollar house running 2% profit and change nothing through 2026. Beef climbs 7.5% at retail and 9.4% wholesale per USDA ERS, other food 3.2%, non-alcoholic beverages 5.7%. With food cost at 31% of sales, that basket adds roughly 1.5 points of cost; if labor moves a point on minimum-wage pressure —16.50 dollars per hour in California, 11.00 for tipped staff in New York per RBT CPAs— prime cost crosses 66% and 24 thousand dollars of profit becomes a loss of about 6 thousand. No crisis required: one quiet year does it. Now flip the exercise, close three points of variance plus two dollars of average check, and the same house finishes the year at 5% margin. What separated the two scenarios was never the market; it was whether somebody costed the plate before the kitchen produced it.

Chapter 9 — Where the two roads genuinely diverge

Cutting has a floor; yield does not. You can bring food cost from 34% to 31%, and past that point you are simply selling a worse product. Server-hour productivity, by contrast, allows sustained improvement: the same room, the same menu and the same hours perform better once the team knows what to recommend and when to stay quiet. The traditional approach confuses input cost with plate cost. Actual cost includes waste, cut yield, inconsistent portioning and shrinkage; the difference between what the recipe says and what the register records is food cost variance, and that is where the missing money lives. A flat menu increase destroys menu engineering. Adding 8% across the board makes the star plate —already converting— more expensive while leaving the dog untouched, still occupying menu space and kitchen minutes without contributing. Traditional operators present sales to their board; mature operators present prime cost, EBITDA per unit and margin per hour worked.

Chapter 10 — Where the two roads genuinely diverge — in practice

With all food forecast at +3.2% for 2026 per USDA ERS, the investor conversation is no longer won by showing top-line growth. Staff turnover is disguised food cost. Every new server sent to the floor without structured training produces order errors, comps and low tickets during those first weeks, and that cost lands on the food line, not on payroll.