Menu engineering: the numbers before and after you touch the menu

The 20% of the menu that pays the payroll

Fourteen dishes with zero sales in a quarter and three with negative contribution margin: that was the finding at the Peruvian kitchen in Chapinero billing 92 million pesos a month across 51 items with a declared food cost of 34%. Toast, in its Restaurant Trends 2026 covering more than 100,000 locations, reports that roughly 20% of dishes generate 71% of sales, and when that split shows up in your POS the remaining 80% is not variety, it is financed inventory. The operational reading allows no nuance: every dead item drags along at least one exclusive ingredient, a weekly waste line and a ticket time the kitchen pays for at peak. Before touching a single price, sort 90 days of sales detail from highest to lowest and mark where the cumulative 71% falls. The right unit is contribution margin in money per dish sold, not food cost percentage, and that distinction is worth between 4 and 8 margin points in 90 days.

Food cost percentage or pesos per dish sold

The National Restaurant Association puts median full-service food cost at 32.0% of sales in 2024, with an optimal band between 28% and 35% and 31.0% among locations above two million dollars in sales; those are range references, not per-dish targets. A ceviche at 33% leaving 21,000 pesos brings more cash than a pasta at 24% leaving 9,400, and sorting the menu by percentage pushes the server to sell exactly the wrong thing. Rebuild your matrix with two columns —units sold and contribution in money— and half of what you called stars will turn out to be something else. A 34% that reads as normal in fine dining is a leak in a QSR. The National Restaurant Association splits the bands by concept this way: 25% to 30% in quick service, 30% to 34% in casual and 34% to 40% in fine dining, with a 32.4% median in limited service during 2024.

The food cost band belongs to the concept, not to the wish

The mistake that repeats most in diagnostics is setting one 30% target for an entire casual dining menu and then punishing the chef over the ceviche, which structurally sits above it. Diego F. Parra and the Masterestaurant framework work the other way around: the band is set per dish family, two or three anchor items are allowed to live at 38% because they bring traffic, and sides and beverages compensate. Put the band per family on the spec sheet, not in the board meeting. There is one menu lever that costs nothing in ingredients: the typography of the price. The study by Cornell University, School of Hotel Administration (2009) measured an 8.15% rise in spending per person when the menu removes the dollar sign and leaves the number alone. Applied to the 92 million pesos a month of the Chapinero Peruvian spot, that is close to 7.5 million a month with no recipe change, no price increase and nobody to train.

Dropping the currency sign moves the check without touching the recipe

It is not magic either: the effect dilutes if the menu still carries 51 items, because a saturated guest retreats to whatever is familiar and cheap. Cut first, lay out second. And if your menu lives in a downloadable PDF, the change costs an afternoon of design and pays itself back that same weekend. The third difference lives outside the dining room, inside the app. A 51-item menu on Rappi with half the photos missing lands below a competitor running 22 complete items, because the algorithm rewards the finished listing and conversion per impression, not catalog size. Circana measured consumer restaurant spending in the United States growing barely 2% in 2024 with flat traffic, and 3% year over year in food and beverage during the first half of 2025: there is no new demand to share out, everyone fights over the same plate. Taco Bell, for its part, reports 20% higher spending through its digital self-service system than at the human register.

The digital channel punishes a long catalog

Bring your digital menu down to what you can photograph well and describe in two lines. Removing dishes raises sales, and that sounds like a contradiction until you watch the kitchen on a Friday at nine at night. Every extra item adds a decision point for the guest and a bottleneck on the line; when the menu drops from 51 to around 28 dishes, ticket times fall, weekly waste concentrates on fewer ingredients and the server can recommend from memory. What would happen if you cut without reading the mix? You would lose the three traffic drivers that fill tables even on thin margin, and sales would sink for a full quarter before recovering. That is why the cut goes by quadrant, never by percentage: out with what neither sells nor earns, redesign what sells badly, and move the stars up the layout. Benchmarks apply differently depending on size, and using them wholesale is the short road to the wrong diagnosis.

How to read these numbers in YOUR operation?

In a small venue of 30 to 40 seats with a 25-dish menu, the goal is not the 32% median but identifying the four dishes that make 60% of sales and locking down their spec sheets.

In a mid-size operation of 60 to 100 seats across two shifts, it pays to measure contribution margin per kitchen hour, because the bottleneck is time rather than cost, and there the 31.0% reported by the National Restaurant Association for locations above two million dollars does become reachable through purchasing volume. In a group of three or more venues, the engineering runs on a consolidated matrix and executes venue by venue, with prices differentiated by area. Two layers usually get mixed and deserve separating. The food cost benchmarks —32.0% median in full service, 32.4% in limited service, bands from 25% to 40% by concept— come from the National Restaurant Association Restaurant Operations Report 2025, with 2024 data and a United States sample; the dollar sign figure is from Cornell (2009) in one contained restaurant experiment; the spending numbers belong to Circana.

Where these numbers come from and what they do not say?

None of those sources measured a 60-seat Colombian restaurant, so the band works as an order-of-magnitude reference and never as a contractual target.

The second layer —what to do with that figure on a Tuesday morning— is consulting judgment from Diego F. Parra and the Masterestaurant framework, and it is presented as such: expert reading of public data, with no proprietary sample behind it. The first difference is one of OBJECT: a menu without engineering tries to please everybody, an engineered menu manages a mix. When 20% of the dishes drive 71% of sales —the split Toast reports in its 2026 Restaurant Trends across more than 100,000 locations— leaving the other 80% untouched means financing, with inventory and labor and waste, a catalog the guest already rejected. The second is a unit-of-measure problem. Most owners look at food cost percentage per dish; the consultant looks at margin dollars per dish sold.

Four differences that actually move cash

A ceviche at 33% that leaves 21,000 pesos beats a pasta at 24% that leaves 9,400, and sorting the menu by percentage pushes the guest in exactly the wrong direction for the till. The third difference lives outside the restaurant, inside the local digital engine. A short, complete menu ranks better on the delivery apps because incomplete listings and long prep times get penalized; that same menu, loaded as products on the Google Business Profile, feeds Maps answers when somebody searches «Peruvian restaurant near me» at 8 p.m. The fourth is CADENCE, and it is the one most people skip. Menu engineering is not a project, it is a 40-minute monthly routine with the item sales report in front of you. Diego F. Parra hammers this point in Masterestaurant audits: a menu redesigned and then abandoned for nine months drifts back to the food cost it came from, because supplier prices do not wait.