Menu Engineering: The Mistake That Costs 12 Points of Margin vs. the Right Method

Why isn't food cost per dish enough to evaluate profitability?

Food cost per dish only measures a percentage; real profitability is counted in dollars of gross margin, and the two rarely line up. A $9 ceviche at 18% food cost earns $7.38 of margin;

a $16 pasta dish at 32% food cost earns $10.88 — nearly 50% more real cash in the register despite doubling the percentage. The pasta wins, no argument, because nobody deposits percentages at the bank. Seven out of ten owners who bring their menu in for review arrive convinced the dish with the lowest food cost is their crown jewel, and they're wrong almost every time. The right question is never how much a dish costs as a percentage — it's how much money it earns, multiplied by how many times it sells. That crossover, dollar margin against real volume, is the only one that reveals whether a dish actually carries the business or just decorates the menu.

How do you correctly classify a dish in the menu engineering matrix?

Classifying a dish correctly means crossing two variables at once: gross dollar margin per unit and units sold over the last 90 days. That crossover produces four quadrants:

star (high margin, high volume), plow horse (high margin, low volume), puzzle (low margin, high volume) and dog (low margin, low volume). A dish at 22% food cost selling just 3 units per service can land in the dog quadrant if its $8 price leaves only $6.24 of margin, while other items generate $12 or more for the same kitchen effort. We require at least 60 days of POS history before assigning a quadrant, since the first four weeks carry seasonal noise that skews any reading. Sorting dishes by food cost alone, without volume, eliminates stars disguised as puzzles and protects dogs that look efficient on paper while quietly draining the register. Moving a single item on the menu requires, at minimum, 60 days of sales per dish broken down by channel — à la carte, combo, delivery, executive lunch — and by shift, lunch against dinner.

What POS data do I need before redesigning the menu?

I've watched design agencies place the least profitable dish in the top-right corner, the highest-attention zone, simply because 'it looks good.' The result is predictable:

food cost doesn't improve, and average ticket drops 4% to 7% in the 60 days after launch. The right process flips that order — matrix with POS data first, graphic design second — and without that history, any redesign is expensive decoration, never a margin strategy. The brief our designer receives already carries each dish's exact placement by quadrant, not the chef's or owner's aesthetic taste. A well-placed decoy dish raises average ticket between 6% and 11%, according to the A/B tests run in restaurants across Bogotá and Mexico City. The mechanism is price anchoring: an $18 grilled steak reads as expensive when a guest evaluates it alone, nothing nearby to measure it against; set beside a $32 premium cut, that same $18 steak suddenly reads as reasonable.

How much does the average ticket increase with a well-placed anchor dish?

Here's what's counterintuitive: the decoy barely needs to sell — if it moved fast, it would stop doing its job and just become another dish to classify by margin and volume.

Ideally it turns over 3 to 5 units per service, no more. I place at least one decoy per menu category, priced 60% or more above the star item in that section, because only then does the anchoring effect hold up in the numbers instead of the owner's gut feeling. Target food cost shifts by quadrant, because each dish type plays a different role in the register. Stars, high margin and high volume, tolerate food cost up to 30% as long as the price sustains $9 or more of gross margin per dish. Plow horses, high margin and low volume, need to stay under 25% to earn their spot on the menu without leaning on volume. Puzzles, low margin and high volume, demand the most work: cutting food cost 4 to 6 points over a 90-day cycle through recipe engineering or a $1 to $2 price adjustment.

What target food cost should I assign to each quadrant of the matrix?

Dogs that don't move up a quadrant across two straight cycles come off the menu, no exceptions.

For us, 32% food cost on any single dish isn't a target worth aiming for — it's the hard ceiling, and nothing crosses it. Real margin at a full-service restaurant isn't decided by standalone dishes — it's decided by combos and delivery, which together account for 30% to 40% of average ticket. Look at a sirloin steak in isolation and it reads as a low-volume puzzle, a candidate for the chopping block; look at it inside a couples combo with two drinks, where 60% of its sales actually land, and its margin profile flips. Cut that steak on the isolated number alone, and an owner removes the exact dish holding up the combo's profitability, then reintroduces it months later without understanding the sales dip. That's how 41% of the restaurants we've audited underestimated the real margin of at least 5 dishes.

How do combos and delivery affect the menu engineering analysis?

Delivery compounds the problem: platform commission pushes real food cost up 3 to 9 points, turning a star into a puzzle on the aggregator if nobody adjusts the price in time.

Reviewing the menu once a year stopped being enough a long time ago: the minimum now is 90 days, the time it takes POS data to become statistically reliable. That window carries enough volume to catch dishes that shifted quadrants from seasonality, from ingredient cost swings, or from moves by nearby competitors. The rule I apply is simple and leaves no loose exceptions: a dish sitting in the dog quadrant for two straight 90-day cycles comes off the menu, unless it serves a documented decoy function. Protein costs alone can move up to 18% in a single quarter, something that happens regularly with beef cuts and seafood, so a dish that was a star in January can easily turn into a puzzle by April if nobody touched the price or the recipe.

How often should I review and update the menu with engineering analysis?

Only quarterly review protects margin in real time; the annual version just confirms, months late, what's already been lost.

Pulling the POS sales report per dish for the last 90 days, broken down by channel and by shift, is the first concrete step, and the one that really allows no shortcuts. If the system doesn't offer that breakdown, configuring it is step zero, before touching a single number. With the report in hand, calculate each item's gross margin in dollars — selling price minus recipe cost — then rank the list from highest to lowest margin and cross it against unit volume sold. Dishes landing in the upper-right quadrant, high margin and high volume, are the stars, and the menu needs to highlight them in the highest-attention zones. The whole process takes 4 to 8 hours the first time on a 40-item menu, and leaves behind an action map with 3 to 5 pricing or recipe decisions that typically move total gross margin 8% to 15% in the next cycle.

The 6 differences that most impact margin

The unit of measurement is the first thing that changes: the mistake counts in percentage, the right method in gross dollar margin per dish and per service. Sales volume never enters the equation in the mistake; the right method always crosses it against margin, anchored in 90 days of real sales. Treating every dish the same is the mistake's signature. We assign a different food cost target by quadrant instead, between 18% and 32%. When aesthetics call the shots, that's the mistake; when POS data and visual-attention zones call the shots, that's the right method. Reviewing the menu once a year is the habit behind the mistake, and whole cycles of silent loss slide by because of it. Every 90 days, instead, dogs that haven't moved quadrants in two cycles get caught and cut. The mistake looks at standalone dishes and hides up to 9 points of real food cost; the right method uncovers it because it crosses sales by channel — à la carte, combo, delivery.