Star, plowhorse, dog and puzzle dishes: which method fits your operation in 2026

What is the best way to classify stars, plowhorses, dogs and puzzles in 2026?

For the independent operator with 20 to 60 tables, a 30 to 60 item menu and nobody assigned full time to analysis, the best option is the Masterestaurant method of menu engineering:

contribution margin in DOLLARS and a rolling 90-day sales mix, not the Kasavana matrix driven by percentage food cost. The reason is treasury, not textbook. The National Restaurant Association placed the median operating margin of independent US restaurants between 3% and 5% in its State of the Industry 2026, and inside a two-point band you cannot afford to sort a menu with a ratio that does not pay suppliers. Percentage food cost is a quotient; contribution margin is money entering the drawer. A risotto at 31% ingredient cost and 11 USD of margin funds more payroll than a salad at 22% and 4 USD, and the classic matrix, as it gets applied in most kitchens, ranks them exactly backwards.

What a DOG really is (and why almost nobody measures it right)?

A dog is not simply the slowest seller: it is the dish that, ON TOP OF selling poorly, delivers a contribution margin below the average of its own category, and that double condition gets lost whenever the menu audit runs on percentages.

Michael Kasavana and Donald Smith published the framework in 1982 and the idea remains intellectually sound: crossing popularity with profitability was excellent thinking. What aged badly is the application, forty-four years later, with restaurant menu prices rising 4,1% in 2024 and 3,8% in 2025 according to the USDA Economic Research Service Food Price Outlook, and with menus split across dining room, delivery and counter. Run a 40-item menu through food cost percentage and you will kill dishes that fund your payroll while protecting dishes that merely look cheap on the spreadsheet. If your operation runs between 20 and 60 tables and nobody spends a full shift studying the menu, use a rolling 90-day window recalculated every month, with margin expressed in dollars per dish.

Best for 20-60 table operations: dollar margin and a rolling 90-day window

An annual classification describes a restaurant that no longer exists. Follow the scenario all the way through: if protein cost moved double digits between quarters — and in 2024 and 2025 it did — half of your January stars are plowhorses by September, you keep pushing them on the chalkboard at the door, every extra sale drains cash instead of adding it, and you find out at the December close, when there is no quarter left to rescue. The 90-day window absorbs seasonal noise without hiding cost drift, and a monthly recalculation fits on one sheet a head chef fills in forty minutes. At 22% average aggregator commission, a 14 USD dish carrying 40% ingredient cost leaves roughly 2,3 USD of margin on delivery versus about 8,4 USD in the dining room, so classifying both channels inside one matrix produces decisions close to random. When more than 20% of your sales leave through a platform, you need two boards, not one.

Best for aggregator-heavy menus: classify each channel separately

Look at how the chains solve it: McDonald's launched McValue in 2025 with bundles that save 15% against buying the components separately, and that package engineering exists because the margin of the set defends itself better than the margin of a loose item. As Diego F. Parra, restaurant consultant and founder of Masterestaurant, explains, the same dish can be a star in the room and a dog on delivery, and treating both cases with a single label is the fastest way to empty the drawer while sales climb. Three scenarios turn the classic Kasavana matrix — the popular option, the one every school teaches — into the worst choice for your restaurant. First: a short menu under 15 items, where each quadrant holds two or three dishes and the popularity average loses statistical meaning; decide dish by dish on absolute margin instead.

When NOT to choose the popular option?

Second: operations where alcohol carries real weight, a category that 46% of Technomic respondents name among the highest-margin lines on the menu according to Nation's Restaurant News (2024);

mixing drink and food inside one matrix shifts the profitability axis and turns perfectly healthy food items into dogs. Third: any menu selling more than 30% through aggregators. In all three the matrix is not wrong in theory, it is badly fed, and a badly fed matrix issues orders with the same confidence as a well fed one. Four signals tell you the method being sold to you will not survive your kitchen. One: if the tool asks for percentage food cost and gives you nowhere to enter dollar margin, you are buying a ratio dressed up as a decision. Two: if the model loads payroll, rent and utilities onto the plate, it will label anything with a long process a dog; those costs belong to break-even, not to the recipe cost sheet.

RED FLAGS when comparing menu engineering methods

Three: if the classification runs once a year, it was born expired against a menu index that climbed 4,1% in 2024 (USDA ERS). Four, and this one costs the most: if nobody asks which channel each unit sold through. A consultant who hands you four quadrants without asking about your delivery mix did not audit your menu, he handed you a template with his logo on top. The puzzle — strong margin, weak turnover — is the quadrant where the most money sits untouched, and also where the standard reaction, cutting the price, gets it wrong. Cutting the price of a puzzle destroys the one good thing it had. What works in an independent operation is moving its position and its wording before touching the price: relocate it to the top third of its section, give it a concrete name, and train the suggestion on the floor. The visibility lever is measurable well outside fine dining: McDonald's reports sales lifts of 5% to 6% after installing self-service kiosks, and the mechanism there is orderly exposure of the product, not price.

What to do with PUZZLES: high margin, low volume?

If after 60 days of relocation and active suggestion the volume has not moved, then pull it, because a puzzle that stalls two months straight is a dog with a pretty cost sheet.

Take a 45-item menu in a 40-table restaurant with a 24 USD average check and 6.000 covers a month. If the eight most promoted dishes were chosen for low food cost and their average margin is 5,80 USD, while six others carrying 9,40 USD of margin sit trapped in the last third of the menu, shifting 15 points of the mix toward those six moves roughly 3.200 USD of monthly contribution without selling one extra plate or raising one price. Against a median operating margin of 3% to 5% (National Restaurant Association, State of the Industry 2026), that number is no fine tuning: it is the difference between closing the year in the black and explaining to your bank why you did not.

The practical case: what changes in the till when you sort by margin

And the whole exercise fits in a spreadsheet with three columns: dish, units sold over 90 days, margin in dollars. The unit of measure. The traditional method asks what share of the price goes to ingredients; the Masterestaurant method asks how many dollars stay on the table when the plate leaves. A risotto at 31% food cost with an 11 USD margin funds more payroll than a salad at 22% with 4 USD, and the classic matrix ranks them backwards. The frequency. An annual classification describes a restaurant that no longer exists: when protein cost moves double digits between quarters, half your January stars are September plowhorses and nobody finds out until year-end close. Channel treatment. At 22% average aggregator commission, a 14 USD plate at 40% food cost leaves 2.3 USD on delivery against 8.4 USD in the dining room. Ranking both channels together produces decisions that destroy margin while the sales report climbs.

Where the two paths genuinely split?

The starting point. The MR method refuses to classify before standard recipe and cost per portion are closed, because a matrix built on estimated costs sorts the error very elegantly.

The traditional method, as commonly practised, tolerates estimation. What happens to the dog. The textbook says pull it. We draw a line: a dog that drags traffic or covers a dietary need for the whole table gets reformulated and repriced against a twelve-week deadline; a dog that merely occupies a menu line and kitchen minutes goes without ceremony.