Menu engineering: stars, plough-horses, puzzles and dogs

Ask a chef in Dubai which dish sells the most and the answer arrives in a second. Ask which dish pays the most and the kitchen goes quiet. They are almost never the same plate, and the distance between them is where the month's profit hides. A full Friday running the wrong mix can bank less than a slow Tuesday running the right one.

What is menu engineering?

Menu engineering is the practice of scoring every dish on two numbers: how many people order it, and how many dirhams it leaves behind after food cost. The menu is then rearranged so the plates that do well on both get seen first. Michael Kasavana and Donald Smith at Michigan State gave the four resulting groups their names in 1982, and kitchens still use them: stars, plough-horses, puzzles and dogs.

The number that decides everything is contribution margin in cash, not food cost as a percentage. A wrap priced at AED 24 on a 25% food cost leaves AED 18 behind. A mixed grill at AED 78 on a 38% food cost leaves AED 48. The grill looks worse on the percentage report every accountant prints, and it pays for nearly three wraps.

What are stars, plough-horses, puzzles and dogs?

Stars sell often and pay well. Plough-horses sell often and pay badly. Puzzles pay well and sell rarely. Dogs do neither, and they are usually the longest section of the menu.

Those four boxes come from two cut-lines drawn across your own sales, not from a rule someone else published. One line separates popular from unpopular. The other separates high cash margin from low. Every dish lands in exactly one box, and the box tells you what to do next. That is the entire point of scoring dishes instead of arguing about favourites.

How do I find the four categories in my own sales data?

Pull item-level units sold for a period long enough to be real (four weeks, or 400 orders, whichever comes second), then put your food cost against each dish. A sales dashboard by day, item and channel gives you the units; an export to a spreadsheet whenever you want it gives you somewhere to type the costs.

Draw the popularity line first. Divide 100 by the number of dishes on the menu to get each dish's fair share, then take 70% of it. Forty dishes means a fair share of 2.5%, so the line sits at 1.75%: anything selling above that share of units is popular, anything below is not.

Draw the margin line second. Add up the total cash margin earned across every dish sold, divide by total units, and you have the weighted average margin per plate. Dishes above it pay well, dishes below it do not. Two lines, four boxes, no opinions.

Three practical warnings. Count units, never revenue, or the expensive dishes drift into the popular half by accident. Exclude the weeks a dish was unavailable; the per-item stock counts that fall as orders come in show you exactly which weeks those were. And run it per branch, because a tower in Business Bay and a high street in Sharjah do not order the same food.

What actually moves the mix?

Four levers move mix reliably: position, naming, add-ons and bundles. Price changes move it too, but they move it slowly and they are the one lever a customer can compare against the shop next door.

Position. Moving an item to the top or the bottom of its block lifts its sales, and the lift is easy to measure on your own numbers within a month. On a phone it is blunter still: a storefront built for one thumb shows six items before anyone scrolls, so the top of a category is the whole shop window. Twelve page layouts and a photo style toggle (square, tall, or none at all) exist because a menu with 9 categories and a menu with 90 dishes need different furniture.

Naming. A name that describes what happened in the pan outsells a noun. "Chicken machboos" is a label; "slow-cooked chicken machboos, dried lime, crisp onion" is an argument. Arabic item names and descriptions sit alongside the English ones, which matters when half a building reads one and half reads the other.

Add-ons. Options and add-ons with their own prices are the highest-margin lines on any menu, because the labour is already paid for. A AED 6 garlic dip that costs AED 1.30 to make adds more cash to a AED 45 grill than a AED 6 price rise would, and nobody compares dip prices. Required choices and limits keep the kitchen able to build what the screen sold.

Bundles. Bundles (any three items for a fixed price) are how a plough-horse drags a puzzle into the cart. Ten kinds of upsell cover the rest: pairings, ladders, thresholds. Discounts never stack by accident, so the best one wins and a bundled cart cannot be couponed down to nothing underneath you.

What do I do with each of the four categories?

Protect stars, fix plough-horses, promote puzzles, and cut dogs unless they earn their place another way. Change one thing at a time so you can tell which change worked.

Stars get the top of the category, the photograph, and no recipe edits. The most common way to lose money in a month is to shave 15 grams off the dish everybody came for.

Plough-horses need margin, quietly. Attach a paid add-on, tighten the garnish, or lift the price by the smallest step the market tolerates. A plough-horse selling 300 covers a month gains more from AED 2 of extra margin than a puzzle gains from doubling its sales.

Puzzles are worth money and nobody sees them. Move them up, rename them, photograph them, and pair them as the suggested side on your best-selling dish. Loyalty helps here more than a discount does: stamps can be set per order, per dirham, or only on the dish you want to sell more of, so the puzzle becomes the plate that earns the reward.

Dogs come off, and the menu gets shorter, which is its own win: fewer lines to prep, less waste, faster tickets. Keep a dog only when it does a job the grid cannot see: the vegan option that lets a table of six choose you, or the dish your name is built on.

Where does restaurant software fit into this?

Restaurant software does not decide which dish is a dog. It removes the two chores that stop most kitchens from ever finding out: assembling clean item-level numbers, and pushing a change live before the idea goes cold.

Every line is re-priced from your menu at checkout, so the price you tested is the price charged and last month's numbers stay comparable. Reordering a category is a drag, not a reprint. The customer gets a WhatsApp confirmation the moment they order, the kitchen gets its ticket, and a weekly cap per customer stops a mix-fixing campaign turning into pestering. VAT invoices carry your TRN whichever way the mix moves.

A menu is the only piece of marketing every single customer reads. Most kitchens spend a year on the recipes and twenty minutes on the order the dishes appear in, then wonder why the wrong plate is winning.

— Anzul Aqeel, founder of Get Menu

There is a commercial reason this work pays better on your own address than on a marketplace. At 25–30% commission, a AED 48 margin becomes a AED 27 margin, and every improvement you engineer is shared with a platform that also keeps the customer. On a flat monthly fee with 0% commission, the margin you win is the margin you keep. The plans are listed on the /pricing page, and the trial runs 15 days without asking for a card.

How often should I run menu engineering?

Monthly, on the same day, against the previous full month. Anything more often and ordinary weekly noise reads as a trend; anything less and a dish that started losing money in October is still on the menu at New Year.

Judge each change on mix share, not on revenue. Revenue moves when the weather changes. Mix share, this dish as a percentage of units sold, moves when your menu changes, which is the only thing you controlled. Set one number to beat, such as lifting average cash margin per plate from AED 21 to AED 24, and stop when you reach it rather than redesigning everything at once.

Frequently asked questions

Do I need a POS to do menu engineering?

No. You need units sold per dish and a food cost per dish, and both can come from an ordering system, a spreadsheet, or a month of counted tickets. A cash register that reports totals but not items is the one setup that blocks you, because the whole method depends on counting plates rather than money.

How many orders do I need before the numbers mean anything?

Roughly 400 orders or four full weeks, and never fewer than about 30 units for any single dish you intend to judge. Below that, one large table skews a whole category. Newer kitchens can still run it — treat the first pass as a shortlist of suspects rather than a verdict, then re-run it a month later.

Should I delete every dog on the menu?

No, but you should be able to say out loud why each one survives. A dog that anchors a dietary need, carries the restaurant's name, or sells the table around it is doing a job the two axes cannot measure. A dog that exists because it has always existed is costing you prep space, stock and menu attention that a puzzle would repay.

Does menu engineering work for a delivery-only kitchen?

It works better, because the menu is a screen you can change this afternoon and the data arrives already itemised. Watch two things print kitchens never see: how far a customer scrolls before ordering, and which add-ons attach to which dish. Delivery also lets you engineer per area, since a delivery fee per building and a free-delivery threshold you choose already vary by tower.