Menu Engineering Explained: How to Design a Menu That Sells Itself

Menu engineering is the habit of judging every dish on two numbers: how often guests order it, and how much cash it leaves behind once the ingredients are paid for. Sort the menu on those two measures and the next decisions get much clearer. You will know what to push, what to reprice and what to quietly retire.

The method is old, and the original idea is credited to Kasavana and Smith. The version below follows the 1995 FIU Hospitality Review paper that extended it. It still works because it needs nothing but your own sales report and your recipe costs.

What does menu engineering actually measure?

It measures popularity and contribution margin for each item, and nothing else. Popularity is the share of a category's sales an item takes. Contribution margin is the selling price minus the food cost of one portion.

The second measure is the one people get wrong. Contribution margin is a sum of money, not a percentage. A dish that sells at 20% food cost and earns little per plate can be worth less to you than one at 40% that earns a lot.

A guest pays you in money, not in percentages.

How do you classify a dish?

You compare each item against two lines: a popularity cut-off and the average contribution margin. The FIU paper treats an item as popular if it sells above 70% of an even share. In a class of eight dishes, an even share is 12.5%, so the cut-off is 8.75% of the mix.

The margin line is the weighted average: total contribution margin divided by total portions sold. An item above it is high, below it is low. That gives four groups, each with a default action from the paper:

Always compare within a category. Mains against mains, desserts against desserts. A dessert will never out-earn a ribeye in cash, and that tells you nothing useful.

What does it look like with real numbers?

Imagine a restaurant with eight mains and 1,000 main courses sold in a typical month. These figures are invented for illustration, not taken from a real business.

Total margin is 17,308, so the weighted average is 17.31 per plate. The popularity cut-off is 87.5 portions. The blended food cost across all eight dishes is 30.0%.

Now look at what a food cost percentage would have told you. The pasta runs at 20%, among the cheapest on the menu (the risotto is 19.1%), yet it is a plowhorse. The ribeye runs at 40.5%, the worst on the menu, yet it is a star. A manager chasing a lower percentage would push the pasta and hide the steak, and cash earned would fall. For more on that trap, see food cost percentage.

What do you do with each group?

Do one thing per group, then watch the sales mix. The paper is clear that the quadrant is not the whole story: raising the price of a plowhorse with a thin margin could push it into the dog or puzzle group if demand falls away.

In the example:

1. Burger (plowhorse). Raise the price by a small step, or trim the garnish cost. Do not remove it, because it brings in the volume that pays for everything else.
2. Pasta (plowhorse). Add a paid extra, such as a protein or a cheese course, rather than raising the base price.
3. Lamb shoulder (puzzle). It earns well but sells slowly. Try a better description, a place near the other mains, or a table-side mention from the team.
4. Vegetable curry (dog). Ask why before cutting. If it is the only vegan main, it may be doing a job the numbers cannot see. Rework it before you remove it.
5. Stars. Leave the recipe and the position alone. Test a small price rise on one at a time.

Pricing methods beyond this are covered in how to price a menu.

What does the matrix miss?

It misses labour, waste, kitchen capacity and the way dishes sell each other. The FIU authors say so themselves. Food-only margin ignores other variable costs. Labour is hard to split into variable and fixed parts. The model also cannot separate two items that land in the same group.

Two corrections help most:

• Labour. A high-margin dish that takes ten minutes of a cook's time on a busy pass may earn less per hour than a plowhorse that goes out in two. Rank items by effort and keep it in mind, as the paper does.
• Dependence. Guests often choose a main first and build the meal around it. A cheap starter that sells a great deal may be there because of a star main, not on its own merit.

The paper also notes that managers paid on food cost percentage have little reason to adopt a cash-margin view. If you run a team, change what you reward before you change the spreadsheet.

Does menu layout really change what guests choose?

Less than the folklore suggests. The popular idea is a sweet spot, a place on the page where the eye lands first. A 2012 eye-tracking study led by Sybil Yang, whose paper analyses 25 participants, did not support it. Readers moved through a two-page mock menu in a book-like order, and no single spot drew the first, last and most frequent looks.

The study also found that diners read the mains most carefully on a first pass and often build the rest of the meal around their main. The same research suggests treating traditional design rules sceptically. A trade report of the work gives no sample size and describes a mock-up menu, so it is one input and no law.

Price presentation has a similar story. The Cornell Chronicle reported that guests at an upscale casual restaurant spent about 8% more when prices were printed as plain numbers without a currency sign. The underlying paper is more cautious: the overall effect of price format on spending was not statistically significant, and the 8% gap comes from one comparison, numerals-only against the other formats combined, in one café over a few months in 2007. Test it before you copy it. Our guide to menu design mistakes covers the practical side.

How do you run this without breaking the menu?

Change one thing per category per cycle, then measure again. If you move three dishes, reprice two and rename a fourth, you will not know which change did the work.

A simple routine:

1. Pull a sales mix report for the period, by item.
2. Cost every recipe fresh. Old costs make the whole exercise wrong.
3. Classify each item within its category.
4. Pick one action for each group and write down what you expect.
5. Review after a full cycle, including a weekend and a quiet midweek.

Keep the language about guests honest. A dish is a dog because of the numbers on a sheet, not because anyone dislikes it. The aim is a menu where the dishes your kitchen cooks best are also the ones that earn their place.

How do you make a puzzle sell without discounting it?

Change how the dish is described and where it sits before you touch the price. A puzzle already earns well, so a discount gives away the very thing that makes it valuable.

Descriptions are the cheapest lever. Work cited in the Cornell price paper (Wansink, Painter and Van Ittersum, 2001) found that descriptive menu labels raised how often dishes were bought and how satisfied guests were, though not what they were willing to pay. That is a useful result for a puzzle: you want more orders at the same price, not the same orders at a lower one.

A few practical moves, in order of effort:

• Name the dish for its method or its source, plainly and truthfully, rather than adding adjectives.
• Place it next to a star in the same category, where guests are already comparing.
• Brief the floor team. A server who has tasted the lamb will describe it better than a line of menu copy.
• Offer a half portion or a tasting size if the full plate feels like a big commitment.

Give each change one full cycle. If the puzzle still does not move, the guests are telling you something the spreadsheet cannot.

What are the most common mistakes, and the fix for each?

The arithmetic rarely fails. The inputs and the follow-through do.

• Stale costs. Recipe costs from last year make every group wrong. Recost the top sellers each cycle, especially anything bought on a volatile market price.
• Mixing categories. A single matrix for the whole menu pushes every dessert into dog territory. Run one per category.
• Counting specials. A dish that ran for two weeks will look like a dog or a star by accident. Leave it out, or compare it only with other specials.
• Cutting too fast. Removing a dog can strand a vegetarian, a child or a guest with an allergy. Check what job the dish does first.
• Changing everything at once. You will not learn which change helped.
• Reading the matrix as a ranking of cooks. It describes a menu on a page, not the skill in the kitchen.

How does this apply to a hotel or a small café?

The logic is the same, but the unit of analysis changes. A hotel breakfast has fewer dishes and a captive audience, so the useful question is often what the buffet or the set menu costs per occupied room. A small café with a short menu can run the exercise on drinks and food separately, since coffee and a pastry behave differently from a lunch plate.

With very few items, the 70% popularity rule becomes blunt. With five dishes in a class, the cut-off is 14% of the mix, and one slow week can flip a group. Use a longer period, or accept that on a short menu you are reading tea leaves, and rely on judgement plus the cost sheet.

What should you do this week?

Pick one category, pull the last two months of item sales, and recost the recipes. Build the table above in a spreadsheet, mark each dish, and choose one action. Write down what you expect to happen, run it for a full cycle, and compare. That is the whole discipline, and it compounds: a menu reviewed four times a year is quietly better every time.

Key takeaways

• Menu engineering sorts dishes by popularity and contribution margin, then picks an action for each group.
• Use cash margin per plate, not food cost percentage. The cheapest dish by percentage can be a plowhorse.
• Compare within categories and recost recipes before every review.
• Add labour and kitchen capacity to the picture, because the matrix leaves them out.
• Treat layout and price-format tricks as hypotheses, and test one change at a time.