Menu engineering: which dishes actually make you money

Menu engineering is a way to decide which dishes to keep, reprice, promote or cut. You sort every dish in one menu category on two questions: does it sell (its share of the category’s sales) and does it earn (its contribution margin, the menu price minus the plate cost, in dollars). The answers put each dish in one of four boxes: stars, plowhorses, puzzles and dogs.

Two thresholds do the sorting. A dish counts as popular when its share of sales is at least 70% of an equal share, so in a category of 8 mains the line is 1 ÷ 8 × 70% = 8.75%. A dish counts as high-margin when its contribution margin beats the category’s weighted average: total contribution margin ÷ total dishes sold.

That is the whole method. This guide covers the data it needs, the math, a worked example with eight mains, what the evidence says about menu design, and what the method cannot see.

What menu engineering is

Menu engineering was developed by Michael Kasavana and Donald Smith at Michigan State University around 1982, according to AHLEI’s menu engineering series, written by Kasavana himself. In short: a menu pays the bills in dollars, not in percentages.

That runs against a common habit: pushing the dishes with the lowest food cost percentage. AHLEI warns that low food cost items are usually the cheaper ones with the smallest contribution margin, so promoting them tends to lower the average check and the gross profit. Food cost percentage still sets a price floor, and our food cost percentage guide covers it. Menu engineering starts once the prices exist and asks which dishes carry the restaurant.

It answers with a two-by-two grid, the menu engineering matrix:

The data you need

Three numbers per dish. A 1997 paper by LeBruto, Ashley and Quain in the International Journal of Contemporary Hospitality Management names them: each item’s selling price, its food cost and the quantity sold.

1. Quantity sold per dish, for one period. Four weeks is a practical length. Whatever length you pick, use it every time.
2. The menu price, before sales tax. If a price changed during the period, use what guests were actually charged.
3. The plate cost. AHLEI’s second article in the series stresses that complete item costs are what make the analysis work, and suggests averaging cost data across a four-week cycle. A stale or guessed plate cost produces a confident, wrong answer, so build yield-adjusted recipe cost cards first.

Analyze one category at a time

Run mains against mains and desserts against desserts. AHLEI recommends splitting items into competing categories, or into meal periods such as lunch and dinner. Mix them and every dessert lands in the low corner simply for being a dessert.

Clean the counts

We found no published standard for these, so treat them as our reasoning:

• Leave out comps and voids. A comped plate adds popularity with no margin behind it. A voided item was never made.
• Decide how extras count. Either fold an add-on’s price and cost into the dish’s averages or leave extras out. Do it the same way every period.
• Compare like periods. A holiday, a private party or a week the fryer was down skews the mix.

The matrix: popularity and contribution margin

Popularity: menu mix % and the 70% rule

Menu mix % is one dish’s share of the dishes sold in its category:

Menu mix % = dish sold ÷ total sold in the category × 100

A dish is high popularity when its menu mix % is at or above 70% of an equal share. AHLEI gives the formula as 1 ÷ N × 70%, where N is the number of items in the category:

Popularity threshold = (1 ÷ number of items) × 70%

The threshold moves with the size of the category, which is why a flat “70%” means nothing on its own:

The sources that set out the rule do not explain why the figure is 70%, so treat it as the method’s convention.

Contribution margin: the weighted average

Each dish’s contribution margin (CM) is its menu price minus its plate cost. The line it has to clear is the category’s weighted average contribution margin:

Total CM of a dish          = CM per plate × number sold
Average CM (the threshold)  = total CM of the category ÷ total dishes sold

A dish above that average is high CM; at or below it, low CM.

Do not take a simple average of the dishes’ margins. It weighs a dish sold 50 times the same as one sold 400 times. The worked example below shows a dish that changes category depending on which average you use.

Stars, plowhorses, puzzles and dogs

AHLEI’s standard actions are short: retain stars, reprice plowhorses, reposition puzzles, remove dogs. In practice each has a few moves.

Stars: popular and high margin

They carry the category. Keep them where guests find them, and protect the recipe, the portion and the supplier. If you raise a star’s price, do it in a small step and watch the next count.

Plowhorses: popular, low margin

Guests like them, but each plate earns less than average:

• Reprice in a small step, then watch the next period’s count before raising it again.
• Re-engineer the plate. Swap the most expensive component for one guests will not miss, or trim a side. Redo the cost card after.
• Sell something with it. A higher-margin side or add-on raises what the order earns.

Do not delete a plowhorse. Guests come for it.

Puzzles: high margin, not popular

Each plate earns well, but too few sell. Make them easier to choose:

• Reposition the dish so it is not buried.
• Rename it or rewrite the description so a guest knows what they are getting.
• Have servers recommend it, and let them taste it first.
• Check the price. If it sits far above everything around it, a small cut may sell enough extra plates to earn more in total.

Dogs: not popular, low margin

The standard answer is to remove them, and a 2022 case study of one Hungarian restaurant also recommends replacing dishes that keep landing there. Before you do, ask one question (our reasoning, not the method’s): does the dish do a job the matrix cannot see? The only vegetarian main, a kids’ plate or a dish that uses trim from another recipe can earn its place. If it does none of those, give its space to something new.

A worked example: eight mains over four weeks

The numbers below are invented for the example, for a US casual restaurant. The salmon and penne carry the same prices and plate costs as in the food cost guide.

Step 1: collect the counts, prices and plate costs

Four weeks of mains, comps and voids removed: 1,500 mains sold across 8 dishes.

Menu mix percentages are rounded to two decimals, so the rows add to 99.99%.

Step 2: the popularity threshold

Equal share          = 1 ÷ 8         = 12.5%
Popularity threshold = 12.5% × 70%   = 8.75%
In dishes            = 1,500 × 8.75% = 131.25

A main needs 132 sales in the four weeks to count as popular. The burger, penne, salmon and short rib clear it. The risotto, at 110, is 22 plates short.

Step 3: the contribution margin threshold

Average CM = total CM ÷ total sold
           = $21,503.60 ÷ 1,500
           = $14.34

The ribeye ($21.50), short rib ($18.60), salmon ($16.51) and risotto ($14.80) are above it. The penne ($13.40), fish and chips ($12.10), burger ($11.20) and grain bowl ($10.90) are below.

Now take the simple average of the eight margins instead: $119.01 ÷ 8 = $14.88. On that line the risotto, at $14.80, falls below average and becomes a dog. The burger and penne sell so many plates that they pull the weighted average down, which is exactly what the threshold should reflect. With the wrong average, a promising dish gets cut.

Step 4: what food cost percentage would have said

By food cost percentage, the penne looks like the best dish on the list and the ribeye the worst. In dollars, every ribeye plate leaves $8.10 more than a penne plate.

Step 5: decide, dish by dish

• Grilled salmon plate, star. Keep it as it is. Hold the portion and the recipe card.
• Braised short rib, star. Keep it, and leave the price alone this period.
• Classic cheeseburger, plowhorse. The best seller, $3.14 below the average margin. A $1.00 increase to $17.00 lifts its margin to $12.20; if it still sells 420, that is $420.00 more over four weeks. Cheaper cheese or a smaller side of fries are the plate-side options.
• Chicken penne, plowhorse. $0.94 below average. A $1.00 increase puts its margin at $14.40. That clears this period’s $14.34, but if both price rises hold at the same counts the average itself rises to $14.82, so the penne would still be a plowhorse: the increase earns $300.00 more, not a new category.
• Ribeye steak, puzzle. The highest margin and 70 sold. Move it up the page, rewrite the description and brief the servers. Keep the price: guests expect a steak to cost more than the mains around it, and the margin is the point.
• Mushroom risotto, puzzle. 22 plates from popular. Have servers suggest it to guests who want a meat-free main, and give it a clearer name and description.
• Fish and chips, dog. $2.24 below average and 90 sold. If nothing else needs the fish, replace it.
• Vegan grain bowl, dog. 50 sold and the lowest margin, but the only vegan main: a table of four with one vegan guest may go elsewhere without it. Rework it with a lower-cost grain or a price closer to the risotto’s.

Then run the same analysis on the next four weeks to see whether the changes worked.

A worksheet to copy

One row per dish, one sheet per category.

Using the printed menu: what the evidence supports

The menu itself is the obvious place to push a puzzle, and menu design is the part of this subject with the most confident advice and the least evidence behind it.

Placement and the “sweet spot”

The idea that a guest’s eye goes first to a spot just above the middle of the right-hand page goes back to scan-path research published in a Cornell quarterly in 1978, according to Nation’s Restaurant News. An eye-tracking study by Sybil Yang of San Francisco State University, published in the International Journal of Hospitality Management in 2012, found no evidence of a sweet spot: on average, diners read the menu in order, like a book. It tracked 25 people and measured where they looked, not what they ordered, so it is moderate evidence. It is still enough to say the “golden triangle” is not established.

Rearranging a real menu has not shown much either. Kincaid and Corsun swapped pages 2 and 3 of a three-page menu in one restaurant and tracked item sales for four months: no significant differences. One restaurant and one change, so moderate evidence.

What that leaves: put a puzzle where a guest reading top to bottom will not miss it, then check the next period’s count. Your own before-and-after run beats any placement rule.

Price formatting

A 2009 Cornell study by Yang, Kimes and Sessarego tested three price formats at lunch in one restaurant, St. Andrews Cafe at the Culinary Institute of America: “$20.00”, “20.” and prices written out in words. Overall, price format was not a significant predictor of spending. In one planned comparison, tables given the numeral-only menu spent more than the other two groups, and the authors note the results may apply only to lunch at that restaurant. Treat dropping the dollar sign as a cheap thing to try, not a proven lift.

Descriptions

A well-known line of research on descriptive dish names came from Brian Wansink, who later ran a lab at Cornell, including a 2001 paper in a Cornell journal on descriptive menu labels and sales. In 2018, Cornell found that Wansink had committed research misconduct, including misreporting research data. Treat his food-labeling findings, and any sales lift quoted from them, with caution. The practical case for a good description needs no statistic: a guest cannot order a dish they cannot picture.

Decoys, boxes and price columns

Decoy pricing, a very expensive item placed to make the next one look reasonable, has been tested mostly in online experiments with hypothetical choices, and the results disagree. One study found an effect on stated choices; another found none. For boxing a dish, breaking up the price column or the ideal number of items per category, we found no reliable evidence either way. Judge them by your own counts.

What menu engineering misses

The matrix is a good first cut, not a verdict.

• Labor. Contribution margin here uses food cost only, so a dish that takes a cook twelve minutes looks as good as one that takes two. LeBruto, Ashley and Quain proposed adding a high or low labor label to each dish in 1995, and in their 1997 paper cautioned that menu engineering on its own is not enough.
• Kitchen capacity. A star that ties up the grill on a Friday can slow every other ticket. The matrix sees plates sold, not minutes at peak. That is our reasoning, and easy to check: ask the line which dishes back them up.
• Substitution. Guests trade between dishes. Raise a plowhorse’s price and some may switch to a cheaper dish rather than pay. Noone and Cachia, using two years of data from 48 outlets of a US steakhouse chain, proposed accounting for these substitutes in pricing and placement decisions (2020).
• Small samples. A dish that sells 30 a month can swing between categories on one large booking. With low counts, use a longer period and do not act on a single run.

Other models

Kasavana and Smith were not first. Miller’s 1980 model sorted dishes by popularity and food cost, and Kasavana and Smith replaced food cost with contribution margin, according to the literature review in a 2022 case study published in Deturope. Pavesic’s cost/margin analysis (1983) sorts dishes by what they cost to produce and what they contribute. Each model uses its own category names, so do not mix one model’s labels with another’s.

How often to run it

AHLEI’s guidance is to revise the menu twice a year and track results quarterly. Between revisions, a four-week run after any price or menu change shows whether the change worked.

Where Platevio fits (and where it does not)

Plainly: Platevio does not do menu engineering. It has no item analytics and no sales-mix reports. It does not total quantities sold by dish or by period, does not calculate food cost and does no margin math. The matrix, the thresholds and the worksheet above happen in a spreadsheet or on paper.

Platevio is a browser-based restaurant POS for orders, the kitchen, the floor and the bill. A few things in it help with the work without doing it for you:

• A record of every order. Each order is kept with its items and extras, as sent to the kitchen and as paid, readable from any browser, anywhere. That is where you read what sold. The counting is yours.
• Prices you can change from any device. When the matrix says a plowhorse needs a dollar more, anyone with permission can update the price, rename a dish or rewrite its description from a phone, tablet or laptop.
• The price is copied onto the order. Each order keeps the price it was ordered at, so a price change mid-service does not change what open orders already cost.

What it will not do: no scheduled prices or promotions, so every price change is made by hand, and no menu photo upload. If the rest fits how you work, Platevio’s pricing is short: Free. No card, no plan.