Dark kitchen menu design: the myth of forty items against the arithmetic of margin

How many items can a dark kitchen actually carry?

Between 12 and 18 items, and the ceiling is set by your walk-in, not by commercial ambition.

Aggregated menu-design research puts the sweet spot at 7 to 15 items per category before decision paralysis kicks in (NeatMenu, Menu Psychology 2026), and a delivery-only kitchen running three categories already hits that ceiling at fifteen dishes. Add the 109 seconds a diner spends on average reading a menu (NeatMenu 2026) and you will see the problem is time, not taste: forty-three items do not fit into ninety seconds of reading, so the customer orders whatever looks familiar. What looks familiar is usually the top 20%. Cutting down to eighteen does not cost you sales; it hands back cold-storage rotation and a purchase list that fits on one page.

The app price is not your revenue: commission eats the costing

Always cost against the net after commission, because the marketplace takes between 18% and 30% of gross depending on country and plan. In cash terms: of every dollar the diner sees on screen, 70 to 82 cents reach your bank account. A 22 USD ticket at 26% commission leaves 16.28 USD real, and those 16.28 are the base you must load with ingredients, packaging and kitchen minutes. If the standard recipe consumes 6.60 USD of raw material, true food cost jumps from an apparent 30% to an effective 40.5%. Eight and a half margin points vanish without any costing sheet recording them, because the sheet was watching the wrong price. That is where the delivery paradox begins: record revenue every month, thinner cash every month, and nobody finds the hole because the hole sits in the denominator.

Sharing ingredients: the 70% that decides whether waste sinks you

Design the menu so at least seven out of ten ingredients appear in more than one dish. This is the only lever that fixes inventory arithmetic in a kitchen without a dining room: when an ingredient turns once a week because it props up a single marginal dish, you are not buying raw material, you are buying future waste. Diego F. Parra orders it this way in Masterestaurant audits: ingredient-dish matrix first, price second, never the reverse. With 43 badly crossed items, a dark kitchen ends up handling fourteen weekly-rotation ingredients; with 16 well-crossed items, that number drops to four or five and the walk-in stops being a graveyard. Theoretical food cost does not move a point, but the real one falls, because waste is what separates paper from cash. Fewer SKUs, faster turns, same perceived menu.

Packaging: 0.85 to 2.40 USD almost no menu absorbs

Packaging is a variable cost per order, not overhead, and it must be assigned dish by dish. Bag, cutlery, sauce cups and container add between 0.85 and 2.40 USD depending on format; a hot bowl with an anti-fog lid costs roughly triple a flat cardboard box. On that 22 USD ticket already down to 16.28 net, a 1.90 USD package is 11.7 extra percentage points of variable cost. Add that to the 40.5% real food cost and you land at 52.2% before touching payroll or rent. A dish can be profitable in the dining room and lose money on the app over this single line, and until you load it into per-portion costing you will not see it. My rule: any item demanding premium packaging pays for it in the price or leaves the menu.

Descriptive names return 12% more and cost nothing

A dish with a descriptive name sells at a 12% premium on average, according to Wansink's research at the Cornell Food & Brand Lab on descriptive menu labels. In a dark kitchen, where the diner smells nothing, sees no pass and has only a compressed photo and three lines of text, that lever is worth more than in the dining room. Apply it to the five or six items already concentrating the bulk of your orders, not to the long tail: lifting a dish that sells 400 units a month by 12% moves cash, while writing poetry for the one selling nine moves nothing. And there is an honest limit worth respecting: the description must match what leaves the window, because a delivery refund is not negotiated with a server, it is settled with a full refund and one star.

Raise prices the way the sector does, not in one jump

The defensible pace is the market's: menu inflation runs at +0.3% monthly in limited service through the first five months of 2026 and +0.2% monthly in full service, according to the National Restaurant Association via Restaurant Business. That drip goes unnoticed; a 9% overnight jump does not. Some 47% of restaurants raised menu prices in the half-year measured by TouchBistro 2024, so adjusting does not put you in a strange position with the customer. What does put you at risk is waiting eighteen months and then correcting all at once. Schedule a review of your fifteen live items every sixty days, move two or three each time —always recalculating on the net after commission, where the app's 26% gets there first— and stop treating price as an annual decision.

Add the category that grows and drop the one that will not turn

Non-alcoholic drinks passed one billion dollars in the United States by the close of 2025 according to Circana, and matcha went from 4.17 billion globally in 2025 toward a projected 7.15 billion by 2030, at an 11.6% CAGR (Grand View Research). These are high-margin categories with cheap packaging and zero complexity at the pass. On the other side, 48.4% of US restaurants already offer plant-based alternatives (Plant Based Foods Association / Datassential 2024), which means that offer stopped being differentiation and became the floor. And watch sesame, the ninth major allergen requiring mandatory declaration in the United States since 2023 under the FASTER Act: in delivery your label is the only warning channel that exists. Swap a stalled item for a high-margin drink before adding number nineteen.

The 3 numbers you should tattoo on yourself

Three numbers and their action, no ornament. First: 26% average app commission, with a real range of 18% to 30%. Action: redo the costing of your fifteen main items today on the net, and pull any that exceed 35% effective food cost. Second: 1.90 USD average packaging per order, within a range of 0.85 to 2.40. Action: assign that cost as its own line in every dish spec and raise the price of items demanding an anti-fog lidded bowl. Third: 109 seconds of menu reading (NeatMenu 2026) against your 12 to 18 items. Action: order the menu so the six dishes concentrating 80% of orders occupy the first screen of the scroll, and push to the bottom those propping up inventory without propping up cash. Start with net costing this week.

Where the math breaks?

The costliest mistake sits not in the standard recipe but in the base it is applied to. Costing per portion against the price published in the app means giving away the commission:

marketplaces take between 18% and 30% of gross depending on country and plan, so what actually reaches your account is 70 to 82 cents on every dollar displayed. Every calculation has to be rebuilt on that net figure, dish by dish, and some items survive in the dining room while dying on delivery for that single line. A second distortion weighs on packaging, and few menus absorb it. Containers, cutlery, single-dose sauces and the bag add between 0.85 and 2.40 USD per order depending on format; a hot bowl with an anti-fog lid costs three times what a flat cardboard box costs.

Where the math breaks — in practice?

When that expense lives outside each dish card, the contribution margin you believe you hold evaporates in the storeroom, and the worst-packing item is usually the one customers order most.

Dishes that hurt profitability are rarely the ones selling least. They are the half-sellers: number nine in the ranking, forty orders a month, forcing three exclusive inputs, one shelf and one plating routine nobody else uses. A dish selling zero gets deleted without drama. The one selling a little while costing a lot defends itself, because there is always a loyal customer asking for it, and that customer costs you 300 USD a month in frozen inventory. Classic Kasavana and Smith menu engineering crosses popularity with margin and hands you four quadrants; a dark kitchen needs one more axis that dining rooms never require, which is the kitchen minute.

Where the math breaks — key points

Two dishes with identical contribution margin are not worth the same when one leaves in three minutes and the other in eleven, because at peak the kitchen does not sell money, it sells minutes, and the slow dish is stealing four tickets from your own register while you celebrate its margin. Comparing this with the dining room is uncomfortable but necessary. At a table, margin recovers through drinks, dessert and suggestive selling, and there a well-built physical menu does work no screen replicates; on delivery there is no server, no recommendation, and the ticket gets built by a thumb scrolling. So restaurant menu design for the digital channel is won in the first four listing positions and in the combo that lifts average ticket without adding a single new input to storage.