Dark kitchen menu design: the costing errors that erase margin, and the method that fixes them

Chapter 1 — What do you actually cost a dark kitchen menu against?

You cost it against contribution margin after the aggregator's commission, never against the dish's food cost.

A tenderloin running 30% food cost leaves 70 points of contribution when it goes out through your own direct-order window, and those same 70 points drop to 43 the moment the aggregator takes its 27% of order value, because commission is charged on the selling price and not on the cost. The arithmetic is merciless and does not negotiate over coffee. When somebody shows me a menu-engineering matrix built on pure food cost, I know they are looking at a business that does not exist: theirs lives with a 15% to 30% intermediation layer glued to every line of the ticket. And with inputs 35% dearer than in 2019 and labor another 35% higher (National Restaurant Association, 2024), the cushion that used to absorb that mistake is already spent. Costing by dish in an operation without a dining room means costing blind, and the reason is that the same product carries two distinct economies depending on where the order comes in.

Chapter 2 — The decision unit is dish-channel, not dish

Take the tenderloin from the paragraph above: 52 points of contribution on a direct order, 31 through an aggregator charging 27%. Twenty-one points of difference that no supplier negotiation will ever hand back. So the operating question stops being «what does this dish cost me?» and becomes «what does this dish leave me BY CHANNEL, and how much volume enters through each one?». If 71% of your orders arrive via aggregator, your real menu is the aggregator menu and the other one is recreational accounting. An operator who grasps this reprices by channel, pushes direct orders with his own incentives, and quits treating the platform as a benevolent partner. Menu prices at large U.S. chains rose 42% between 2020 and 2025 (One Haus), nearly double the 22% general inflation, and much of that gap is precisely this: the channel collecting. A dish with a superb margin that carries 3% of the mix saves absolutely nothing, while one with a mediocre margin and 34% turnover defines your quarterly EBITDA.

Chapter 3 — Sales mix outranks unit margin

That is the whole point of menu engineering and also the reason so many people misapply it: they stare at the unit contribution column, fall for the expensive plate, and neglect the one that pays payroll. Contribution WEIGHTED by turnover is the only figure that matters when you decide what stays on the menu and what disappears. Turn it around for a second. If tomorrow you cut the nine long-tail items that add up to 4% of the mix, what happens? Waste drops, prep time shortens, your stars turn faster, and you probably lose two customers who ordered that odd chicken. You came out ahead. Labeling a dish «Most popular» or «Chef's favorite» lifts orders between 13% and 20% (NeatMenu, 2026), so the mix gets steered rather than endured. Raising your star item 8% can cost you 15% of volume, while that same 8% on an inelastic side dish will not move a single unit.

Chapter 4 — Elasticity: where to raise price and where not to

This is where almost everyone spreads the increase blindly, touching the entire menu with one percentage because it is convenient from the back office. Bad business. Food-away-from-home inflation came in at 3,8% in 2025 (USDA Economic Research Service, 2025) after 4,1% in 2024 (USDA Economic Research Service, 2024), and in Colombia the adjustment reached 9,8% to sustain 98.000 jobs (ACODRES, 2025). Move those points where demand will not feel them: drinks, add-ons, sauces, combos. Alcohol, where licensing allows it, shows up as a top-margin category for 46% of surveyed operators (Technomic / Nation's Restaurant News, 2024). Your star, by contrast, gets handled with tweezers or redesigned from the inside —portion weight, cut, garnish— before anything happens on the outside. A badly designed menu does not punish proportionally to size: it hits differently in each annual revenue band. Below 500.000 dollars, with a kitchen build under 150.000 dollars (Square, 2024), every extra item gets paid for in the owner's hours, so the recommendation is blunt: twelve to sixteen items, two proteins, zero exclusive sauces.

Chapter 5 — The same menu weighs differently by revenue band

Between 500.000 and one million the first head chef arrives and with him the temptation to expand; the brake there is turnover, not appetite. From one to five million the problem migrates to inventory and to food cost variance between shifts, which is where a forty-item menu bleeds without anyone spotting it in the monthly P&L. Above five million the dominant cost stops being the input and becomes coordination: a price matrix by channel, by virtual brand, by city. With each staff departure costing 150% of salary in replacement (StaffedUp, 2025), a complex menu is also a menu that burns cooks. Above five million in revenue, the celebrity-chef restaurant or the large-format themed concept opening dark kitchens plays with costs the small band would not recognize. The menu stops being a list of dishes and turns into a brand license: there are royalties, there is standards auditing, there is product photography costing what a full startup kitchen costs —remember the sub-150.000 dollars of a QSR per Square (2024)— and there is reputational risk an anonymous outlet never runs.

Chapter 6 — High end: when the chef's brand walks into the ghost kitchen

A badly executed plate in a dining room gets fixed by the manager with a glass of wine; in delivery it becomes a two-star review, and each rating star moves between 5% and 9% of revenue according to Michael Luca of Harvard Business School. That is why in this band the menu gets cut MORE, not less: eight to ten armored items, zero tolerance for variation, a price that holds the promise. Chipotle opened between 315 and 345 locations in 2025 with over 80% in drive-thru format, and its menu still fits on a napkin. The sequence we work with at Masterestaurant inverts the usual order, and that inversion is the entire contribution. First you set the target contribution margin per channel after commission; then you derive the price; last, and only last, you ask which dish fits inside that price with the food cost available. Diego F.

Chapter 7 — The Masterestaurant method applied to the dark menu

Parra insists on a hard threshold that stays off the table: maximum 32% food cost per dish, with payroll, rent and utilities NEVER loaded onto the plate, because those live at break-even and pushing them into the recipe card produces prices the market rejects. I will admit that for years I argued the opposite, prorating fixed expenses onto every recipe, and all I achieved was expensive menus whose margins still refused to close. The figure that governs all of this is prime cost, not food cost in isolation. And with labor 35% dearer than in 2019 (National Restaurant Association, 2024), the room for error has vanished. Open your last ninety days of sales, sort items by units sold, and mark the line where the cumulative reaches 80% of volume: everything below it is a candidate for removal. That exercise, which takes two hours, usually reveals that between nine and fourteen items generate under 5% of orders while occupying storage space, mise en place time and the cook's attention.

Chapter 8 — What to do Monday with the menu you already have

Then calculate, dish by dish and channel by channel, the contribution after commission —15% to 30% depending on the platform— and rank them from highest to lowest. What you see in that second column is your real business. One figure to close and open at once: matcha delivery orders grew 34% in 2025 in the United States (Grubhub, 2025 Delivered Report), and whoever caught it early folded it in at a marginal input cost. Margin opportunities show up in the categories nobody watches, almost never in the main course. FIRST, the unit of decision. On a dine-in menu the unit is the dish; on a dark kitchen menu the unit is the dish-channel pair, because the same beef cut yields 52 contribution points on direct orders and 31 through an aggregator charging 27%. Costing without splitting channel is costing blind. SECOND, the weight of sales mix. A dish with a beautiful margin at 3% of the mix rescues nothing; a mediocre margin at 34% of the mix defines your EBITDA.

Chapter 9 — Four differences that decide the margin

Menu engineering exists precisely so you stop looking at dishes and start reading contribution weighted by turnover. THIRD, elasticity. An 8% increase on your star reference can cost you 15% of volume, while the same 8% on an inelastic side item moves not a single unit. Flat increases are the most expensive mistake and the most common one whenever inflation squeezes. FOURTH, the implicit CapEx of variety. Every extra reference drags an exclusive input, a storage slot, a procedure and a waste probability. Forty dishes in 40 square meters is not a generous menu, it is a structural vulnerability wearing a generous name.