Guides

Menu Engineering Beyond the Matrix: What Changes Sales on the Page (2026)

Classifying a dish as a Star, Workhorse, Puzzle or Dog is the easy half of menu engineering. The eye-tracking "golden triangle" consultants sell has been tested against real sales data and failed; descriptive names, price wording and category position have each replicated instead.

Alex Riesenkampff

11 September 2026 · 13 min read · Markdown

Classifying a dish as a Star, Workhorse, Puzzle or Dog tells you which quadrant it sits in. It says nothing about what to do once you're holding a blank menu template and the classification in front of you. The "golden triangle" eye path, one of the most confidently repeated pieces of menu-design advice, has been tested against real point-of-sale data and found to move nothing measurable; boxing an item to draw the eye has never been tested at all, only assumed from that same null result. Three quieter changes have replicated in actual restaurants instead: a descriptive name alone sold 27% more of the identical dish in a Cornell field study, and price wording and category position each moved real orders too. Below is what the evidence supports, what it doesn't, and how to tell within a few weeks whether a change did anything at all.

The "golden triangle" eye-tracking claim has been tested, and it failed

The idea that a diner's eyes travel a fixed path across the page, usually drawn as a triangle landing somewhere in the upper right, and that placing your most profitable dish there sells more of it, is one of the oldest claims in menu consulting. The one field experiment built to test it directly, switching the contents of two full pages on a real three-page menu for four months and comparing sales before and after, found no significant differences in item sales from time 1 to time 2 (Kincaid & Corsun, International Journal of Contemporary Hospitality Management, 2003). The claim traces back further than most people repeating it seem to know. The usual citation is a 1987 Gallup Organization report commissioned by the National Restaurant Association, and later academic reviews of that same report note it established only where diners' eyes tended to land. It never showed that landing there changed what they ordered. Attention isn't a sale. If your current layout is workable, a rebuild around a triangle nobody has shown to move revenue is a project you can skip.

Descriptive names sold 27% more of the same dish

A name costs nothing to reprint, and unlike layout, it has a controlled field test behind it. Brian Wansink and colleagues at Cornell's Food and Brand Lab ran a six-week study in a faculty cafeteria at a Midwestern university, giving identical dishes either a plain name or a descriptive one covering the same food, renaming plain Zucchini Cookies as Grandma's Zucchini Cookies. Items carrying a descriptive name sold 27% more than the identical dish under a plain one, and diners rated the food and the venue more highly afterward, though they weren't willing to pay more for it (Wansink, Painter & Van Ittersum, Cornell Hotel and Restaurant Administration Quarterly, 2001). The willingness-to-pay result matters specifically for a Puzzle-quadrant dish, one with good margin but weak demand: the standard advice is to rename or reposition it rather than cut it, and this is the one part of that advice with a controlled trial behind it. A description naming an origin, a method or a specific ingredient carries more weight than an adjective describing how good the dish is meant to be.

What raised the average check wasn't the dollar sign by itself

Researchers gave 201 real dining parties at a cafe inside the Culinary Institute of America one of three otherwise identical menus: prices printed as "$20," as a bare "20," or spelled out as "twenty dollars." Parties given the numeral-only menu spent $5.55 more per check, about 8%, than parties given either the "$" or the spelled-out format, and the "$" and "twenty dollars" versions performed identically to each other (Yang, Kimes & Sessarego, Cornell School of Hotel Administration, 2009). Sheryl Kimes, the study's co-author and the Singapore Tourism Board Distinguished Professor of Asian Hospitality Management at Cornell, described the mechanism directly: "references to dollars, in words or symbol, reminds people of the 'pain of paying,'" (Cornell Chronicle, 2009). The popular version of this finding, drop the dollar sign, is only half right. The data shows any explicit reference to money correlates with lower spend, symbol or word, so removing the "$" alone while the number itself still reads as a price gets you most of the change for the smallest visual edit. Worth stating plainly: the study's own overall statistical model fell short of significance once every other factor was controlled for, so treat this as an unconfirmed effect worth testing on your own menu, not a proven lift.

Position within a category is the one placement rule that held up

Menu consultants have made a great many claims about where a dish should sit; only one has survived a real field test. Dayan and Bar-Hillel tested position within a category rather than which page a dish appeared on. Their field test alternated a baseline menu with one that moved items to the ends of their own category, in a Tel Aviv cafe over 15 trading days, recording 459 orders on the baseline menu and 492 on the modified one. Moving a dish from the middle of its category to either end raised its share of category orders from 45% to 55%, a relative gain of roughly a fifth (Dayan & Bar-Hillel, Judgment and Decision Making, 2011). The same paper draws the contrast with the golden-triangle claim explicitly: menu consultants "may be wrong in some of their recommendations... but apparently they are not wrong on the particular one studied here." The practical instruction is narrow rather than sweeping: don't spend time studying where the eye lands on the whole page; do put your Star and your renamed Puzzle items first or last within their own section instead of in the middle of six pasta dishes.

Photos help a clear name, and can work against a vague one

A widely repeated claim puts a photo's sales lift at 6.5%, usually credited to "a 2014 Cornell study." That figure has no traceable author, journal or report title behind it anywhere it circulates; every source repeating it cites only other blogs repeating the same number, the pattern of a stat that has drifted away from whatever it was originally attached to. Treat it as unverified. The research that does exist is more conditional than a flat percentage: a 2017 study in the International Journal of Hospitality Management found that a photo helped a dish's evaluation and perceived value most when the dish's own name was already specific. It could work against certain readers' evaluation of the dish when the name was vague instead (Hou, Yang & Sun, 2017). A photo is not a universal lift: the 2017 study found it helped most when the dish's name was already clear, and could work against the dish when the name was vague, so a photo reinforces a name that's already doing its job rather than rescuing one that isn't. If you photograph anything, photograph the dish your description already names precisely rather than the one you're hoping the picture will explain.

Cutting a low-scoring dish can cost more than it saves

None of the above changes a simple operational fact: a dish can be a genuine Dog by the numbers and still be the reason a specific table walked in. Chili's found this out in real time in October 2025, when it pulled its long-running Skillet Queso for a new recipe. The backlash on social media ("You can't take Skillet Queso away, that's like a staple from you guys") was loud enough that within three weeks a spokesperson was already promising "there's still room for two quesos on the menu" (Fox News, 2025). The original recipe didn't return to menus nationwide until December 9, 63 days after it was pulled (FSR Magazine, 2025), which says something on its own: even a chain that capitulates fast can take two months to fix a menu decision it already knows was wrong. A dish's score in the matrix says nothing about whether it's the reason a table booked in the first place, and cutting it can cost the whole party's business, not just that line's own revenue. Larry and Adam Mogelonsky, Partners at Hotel Mogel Consulting Limited, put it plainly: "if you lose the dish then the whole party may end up elsewhere" (Hospitality Net, 2020). Evan Rich, chef-owner of the Michelin-starred Rich Table in San Francisco, prices a 14-ounce dry-aged ribeye at $49, well under what he calls the realistic $58-to-$65 range. He holds the anchor dish below its true cost on purpose, because "it's our job to create value by building a balanced menu that makes sense and works — both for the guest and for a workable business" (San Francisco Standard, 2025). Donnie McClanahan, an independent multi-cafe operator in Knoxville, found a middle path for his own patty melt when its sales dipped. Rather than cutting it, he turned it into a seasonal specialty: "I was able to avoid cutting a menu item that was actually a big hit and instead turned the patty melt into a seasonal specialty" (Square, 2025). Before cutting anything the matrix flags as a Dog, check whether it's doing a job the spreadsheet can't see: anchoring the rest of the menu's pricing, bringing in a specific regular, or carrying the dish that gets your venue recommended by name.

The matrix measures margin, and misses the rest of a menu's job

The four-quadrant matrix from our menu pricing guide classifies a dish on two numbers: how often it sells and how much it contributes when it does. Two hospitality-school researchers argue that misses half of what a menu is doing. "A dish that performs well on paper may still reduce overall profitability if it slows preparation, extends dining duration, or constrains peak-period capacity," says Dr. Cindy Heo, Associate Professor in Revenue Management at École hôtelière de Lausanne (EHL Insights, 2026), naming exactly the kind of cost the matrix's two axes don't capture. Her colleague Dr. Marc Stierand, Full Professor for Service Management at EHL, states the menu's other job in one line: "A menu is a restaurant's business card. It shapes how guests understand the concept, assess value, and interpret the logic of the overall dining experience." A dish can lose on both matrix axes and still earn its place: recipe-costing vendor meez tells its own users that "some dishes stay on menus because they matter to the identity of the restaurant — even if the matrix alone wouldn't justify them" (Liz Van Hoose, meez, 2026). That's alongside every extra item's real cost in prep, purchasing, inventory, training and waste, which the matrix's two axes don't price in either. None of this argues against the matrix; it argues for treating its output as one input rather than the final word on what happens next.

Regulars reward consistency more than novelty

A menu redesign is one pass, not a standing habit, and research on why guests come back says that's the right instinct. Consistency, not new items, is the strongest driver of restaurant loyalty. Claire Tassin, an analyst at Morning Consult, found it in a survey of 4,416 US adults fielded in July 2024: "the most consistent drivers of restaurant loyalty are inherently table stakes for any diner: that the food consistently meets expectations, and it tastes good" (Morning Consult, 2024). New items play a supporting role once that baseline is met, not a replacement for it. Changing a menu too often carries a real cost beyond guest confusion: a new dish has to be trained into a kitchen and a serving team before it runs smoothly, and pulling one before that training pays off wastes the investment already made (GoFoodservice, 2026). Treat this article as a single pass triggered by the matrix, a rename, a reposition and a controlled test, rather than a standing practice of reshuffling the menu every time the data has an opinion. What that consistency is worth in money, rather than in goodwill, is set out in our economics of the repeat guest.

How to tell, in a few weeks, whether it worked

A redesign you can't measure is a guess with better formatting. Change one variable, run it for at least a full week against a like-for-like comparison period, and track category share and contribution margin rather than total revenue: this is a monitoring heuristic, not a controlled trial, and a low-volume dish can still move on ordinary noise rather than your change. No peer-reviewed source has published a validated minimum test window or sample size for a menu change specifically, so treat any precise figure quoted elsewhere as a working rule rather than a law, and treat one week's result on a slow-selling item as a lead to watch rather than a verdict.

A homemade-cakes cafe in Cologne did the low-tech version of this exercise directly. It cut its dessert list down to its five best-selling cakes, dropping the flops rather than trying to reposition or rename every underperformer, and saved €500-700 a month, with less waste and baking stress. That's the kind of change the evidence above supports, cheap, reversible, and measurable within a few weeks, unlike the page-layout work most menu-design advice spends its time on.

Super44 pairs a frontier AI model with hospitality-specific knowledge and your own connected sales data, so instead of guessing which of your Puzzle items deserves a rename, you can ask it directly which dishes are underselling relative to their margin, and then track what happened to that item's sales in the following weeks once you've made a change. The classification behind all of this, and the free calculator that turns your own food cost and volume into a contribution-margin number per dish, is in our menu pricing guide. If a redesign clears the way for a price move on a specific item, our guide to raising prices without losing customers covers what protects guest loyalty once the number on the page changes.