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?

An independent running 20 to 60 tables with no dedicated analyst should use the Masterestaurant method: contribution margin counted in DOLLARS, sales mix recalculated every 90 days, instead of the Kasavana matrix driven by percentage food cost.

The reason is treasury, not textbook. Median operating margin for independent US restaurants sat between 3% and 5% in the State of the Industry 2026, and two points of room leave nobody free to sort a menu by a quotient that suppliers cannot bank. Percentage food cost is a ratio; contribution margin is cash entering the drawer. A risotto at 31% ingredient cost that leaves 11 USD pays more kitchen hours than a salad at 22% that leaves four, and the classic matrix, as most kitchens apply it, ranks that pair exactly backwards.

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

Selling slowly is not enough to condemn a dish.

A dog sells under its family average AND returns less margin than the average of its own category, and the second half of that test vanishes the moment a menu audit runs on percentages. Michael Kasavana and Donald Smith published the framework in 1982; the thinking holds up, since crossing popularity against profitability was a fine idea. What fell behind is the way operators apply it forty-four years later, with menu prices up 4,1% in 2024 and 3,8% in 2025 per the USDA Economic Research Service Food Price Outlook, and with menus split three ways between room, delivery and counter. Run 40 items through food cost percentage and you will kill the plates that fund payroll while protecting the ones that only look cheap in the spreadsheet.

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

A rolling 90-day window, recalculated monthly, with margin stated in dollars per dish, is what holds up an operation running between 20 and 60 tables where nobody studies the menu full time. Annual classification describes a restaurant that no longer exists. Push the scenario to its end: protein cost jumps double digits between quarters — it did in 2024 and 2025 — half your January stars are plowhorses by September, the chalkboard still pushes them, every extra cover drains cash rather than adding it, and the news arrives at the December close, with no quarter left to rescue. Ninety days absorb seasonal noise without hiding cost drift, and the monthly recalculation fits on one sheet a head chef fills in forty minutes.

Best for aggregator-heavy menus: classify each channel separately

Two boards, never one: any operator sending more than 20% of sales through a platform has to split the ranking. The arithmetic decides it. At 22% average commission, a 14 USD dish carrying 40% ingredient cost leaves roughly 2,3 USD on the app and around 8,4 USD at table, so a single grid over both channels is close to a coin toss. Watch how the chains handle it: McDonald's launched McValue in 2025 with bundles that save 15% against the loose components, and that package engineering exists because a set defends its margin better than a single item does. As Diego F. Parra, restaurant consultant and founder of Masterestaurant, explains, one dish can be a star in the room and a dog on delivery; one label for both empties the drawer while sales climb.

When NOT to choose the popular option?

Three situations make me tell an owner to skip the classic Kasavana matrix, the choice every school still teaches. First, a short menu under 15 items:

each quadrant holds two or three dishes, the popularity average stops meaning anything, and the call belongs dish by dish on absolute margin. The second shows up where alcohol carries weight — 46% of Technomic respondents place it among the highest-margin lines on the menu, per Nation's Restaurant News (2024) — because mixing drink and food inside one grid tilts the profitability axis and turns healthy food items into dogs. Third comes any menu selling above 30% through aggregators. Theory is not the problem here. The feeding is, and a badly fed grid issues orders as confidently as a well fed one.

RED FLAGS when comparing menu engineering methods

When the tool being sold to you asks for percentage food cost and offers no field for dollar margin, you are buying a ratio dressed as a decision. That is the first of four signals. Second: a model that loads payroll, rent and utilities onto the plate will call anything with a long process a dog, and those costs belong to break-even, not to the recipe sheet. Third signal, the once-a-year classification, born expired against a menu index that climbed 4,1% in 2024 (USDA ERS). The fourth costs the most: nobody asks which channel each unit sold through. A consultant who hands over four quadrants without asking about your delivery mix audited nothing and simply printed a template with a logo on top.

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

Cutting the price of a puzzle destroys the only good thing it had. That quadrant — strong margin, weak turnover — is where the most money sits untouched, and where the owner's instinct grabs the wrong lever.

In an independent house what works is position and wording ahead of price: move it into the top third of its section, give it a concrete name, train the suggestion on the floor. Visibility is measurable far outside fine dining, since McDonald's reports 5% to 6% sales lifts after installing self-service kiosks, and the mechanism there is orderly exposure rather than discount. Sixty days later, if the relocated and suggested dish still has not moved, pull it: a puzzle stalled two months running is a dog with a pretty cost sheet.

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

Run the numbers with me on a 45-item menu, 40 tables, a 24 USD average check and 6.000 covers a month. Eight dishes promoted for low food cost average 5,80 USD of margin; six others, worth 9,40 USD each, sit trapped in the last third of the card. Shift 15 points of the mix toward those six and monthly contribution moves by roughly 3.200 USD, with no extra plate sold and no price raised. Against a 3% to 5% median operating margin (National Restaurant Association, 2026), that figure is no fine tuning: it decides whether you close the year in the black or explain to your bank why you did not. Three columns hold the whole exercise: dish, units sold over 90 days, margin in dollars.

Where the two paths genuinely split?

Start with the unit of measure. Traditional practice asks what share of the price goes to ingredients; we ask how many dollars stay on the table once the plate leaves the pass.

A dish whose ingredients weigh 31% but returns eleven dollars pays more payroll than one at 22% returning four, and the classic matrix ranks them backwards. Frequency. An annual classification portrays last year's restaurant: protein moves double digits between quarters and, by September, half of January's stars have turned into plowhorses, though nobody notices before the year-end close. Then comes the channel. A 22% average commission turns one plate into two different businesses: 2.3 USD of margin through the app against 8.4 USD in the room, on a 14 USD price at 40% ingredient cost.

Where the two paths genuinely split — in practice?

Rank both channels together and margin dies while the sales report climbs. We refuse to classify anything before the standard recipe and the cost per portion are closed, because a matrix built on estimates sorts the error beautifully.

Traditional practice, as it usually runs, tolerates the estimate. On dogs I part company with the textbook. Pull it, says the school. I split the case in two: a dog that drags traffic or covers a dietary need for the whole party gets reformulated and repriced, with twelve weeks to prove itself; the one that merely occupies a menu line and kitchen minutes goes without ceremony.