Chapter 12: Menu Engineering: Stars, Plowhorses, Puzzles, and Dogs

"The chef wants to pull the trout because the food cost is thirty-seven. The trout pays more rent than the burger. Somebody in this room has to say that out loud." — constructed; the argument that happens in every menu meeting

Overview

Here are two plates from the restaurant we have been building.

One is a trout. It costs $10.40 to put on a plate and sells for $28.00 — a 37.1% food cost, the worst number on the menu, the line the chef apologizes for when you go through the cost cards. The other is a burger. It costs $5.10 and sells for $21.00 — 24.3%, a number you could frame.

Now stop looking at the percentages and look at the money.

The trout leaves $17.60** in the till. The burger leaves **$15.90.

The item with the worst food cost percentage on this menu earns a dollar and seventy cents more per plate than the item with the second-best. That is not a trick, an edge case, or a constructed paradox. It is what happens on every menu where prices were set by a human being rather than by multiplying every plate cost by a constant. And if you manage by percentage — which most operators do, because percentage is the number the software prints in bold — you will pull the trout, push the burger, and be poorer every single week for it.

Chapter 1 said you bank dollars, not percentages. Chapter 11 said it again while it was building the cost cards. This chapter is where you stop nodding at the sentence and start running your menu on it.

Menu engineering is the discipline of treating a menu the way an investor treats a portfolio: not as a list of dishes but as a set of positions, each one earning a certain number of dollars a certain number of times a week, each one a candidate for holding, promoting, repricing, or closing. It gives you two axes — how much a plate earns and how often guests choose it — a two-by-two grid, and four memorable animals to hang on it.

It is also, and this matters more than the trade press admits, a blunt instrument that will confidently tell you to cut the wrong item. So this chapter does both jobs. Sections 12.1 through 12.6 build the tool properly and place all five of Bellwether's dinner entrées on the grid with the arithmetic shown. Section 12.7 spends a long time on everything the grid cannot see — labor, station load, cross-utilization, guest counts, seasonality, sample size, and the uncomfortable fact that both axes are defined relative to the menu you already have, so every menu produces stars and dogs by construction. Section 12.8 tells you how often to run it and, more usefully, what to change first.

In this chapter, you will learn to:

• Compute contribution margin per plate and explain, in dollars, why a low food cost percentage is not the same thing as a profitable dish.
• Compute menu mix percentage and the popularity index for a category, and derive the popularity threshold honestly instead of accepting the one your software picked.
• Build the matrix from your own point-of-sale data, place every item, and show the working.
• Work each of the four quadrants to a decision with a dollar figure attached, including the substitution question the matrix never asks.
• Compute weighted contribution margin for a whole menu and use it to compare a repricing against a mix shift before you do either.
• Say precisely where this tool misleads, and refuse to act on it when the data cannot support the recommendation.

Learning Paths

🏗️ Opening — you have no sales history, so §12.6 and the Business Plan checkpoint are yours: a projected mix is the most fragile number in a business plan, and you need to know how fragile. Read §12.7 twice before you commit to a menu you cannot staff. 📋 Managing — this is your chapter. §12.2, §12.3, and §12.8 are a monthly job you can start next week with data you already own. §12.5 is the four conversations you will have with your chef. 🍸 Beverage — the matrix works on cocktails, by-the-glass wine, and draft lines exactly as it works on entrées, and the contribution margins there are both larger and more spread out. Note in §12.7 that beverage attachment can reverse an entrée's ranking outright. 🚚 Small Format — with six items instead of forty, your popularity threshold is high and your quadrants swing violently on a handful of units. §12.2's sample-size discipline and §12.7's relative-axes problem matter to you more than to anyone else in this book.

12.1 Contribution margin: why you bank dollars, not percentages

Contribution margin (CM) is menu price minus plate cost. That is the whole definition.

$$\text{Contribution margin} = \text{menu price} - \text{plate cost}$$

It is the number of dollars a dish contributes toward everything that is not the food itself: the labor that cooked it, the rent on the room it was eaten in, the gas, the insurance, the loan payment, and — if there is anything after all of that — the profit. Chapter 11 computed it for every plate on Bellwether's dinner menu while it was building the cost cards. Here is why it is the vertical axis of everything that follows.

Food cost percentage answers the question how efficiently did I convert product into price? That is a real question. It is not the question a business asks at the end of the night. The business asks how many dollars did we keep? — and dollars are what pay the rent. A landlord has never once accepted a percentage.

The ranking flips

Take Bellwether's five dinner entrées, priced in Chapter 10 and costed in Chapter 11, and rank them two ways.

FIGURE 12.1 — The same five plates, ranked two ways                  [the Bellwether plan]

  BY FOOD COST %                        BY CONTRIBUTION MARGIN (dollars per plate)
  ──────────────────────────            ──────────────────────────────────────────
  1  squash and grains    16.5%         1  pork chop             $20.91   (was 4th)
  2  Bellwether burger    24.3%         2  Hearth Chicken        $20.48   (was 3rd)
  3  Hearth Chicken       29.4%         3  squash and grains     $20.05   (was 1st)
  4  pork chop            32.5%         4  ember trout           $17.60   (was 5th)
  5  ember trout          37.1%         5  Bellwether burger     $15.90   (was 2nd)

  The two orderings agree about almost nothing. The best food cost percentage on
  the menu is third by dollars. The worst is fourth — ahead of the item ranked
  second on percentage. Only the trout and the burger swap all the way across,
  but that swap is worth $1.70 on every plate, every night, forever.

Read the right-hand column and notice how tightly it is packed. The five contribution margins run from $20.91 down to $15.90 — the entire menu fits inside a band of $5.01, and the top four fit inside $3.31. That is a very flat portfolio, and it changes what menu engineering can and cannot do for this restaurant. Hold that thought; §12.6 collects on it.

🧮 Run the Numbers

The trout and the burger, settled.

	Ember trout	Bellwether burger
Menu price	$28.00 | $21.00
Plate cost	$10.40 | $5.10
Food cost %	37.1%	24.3%
Contribution margin	$17.60** | **$15.90

The trout is 12.8 percentage points worse on food cost and $1.70 better per plate.

Now put volume on it. The trout sells 45 a week; the burger sells 73.

• Trout: 45 × $17.60 = **$792.00 a week, or $41,184 a year**.
• Burger: 73 × $15.90 = **$1,160.70 a week, or $60,356 a year**.

So on the year, the burger out-earns the trout by about $19,000 — because it sells 62% more often, not because its food cost is better. Volume did that. Percentage did nothing.

Here is the decision that actually matters. Suppose you pull the trout and every one of those 45 guests orders a burger instead. You have traded a $17.60 plate for a $15.90 plate, 45 times a week:

$$45 \times \$1.70 = \$76.50 \text{ a week} = \$3,978 \text{ a year, gone}$$

You will have improved your food cost percentage. Your food cost report will look better every month for the rest of the lease. And you will have less money.

Why this keeps happening

The trap is not stupidity. It is that food cost percentage is genuinely useful for a different job. As a control number — comparing what you should have used against what you did use, week over week, as Chapter 11's ideal-versus-actual section showed — it is exactly right, because it holds the menu constant and asks whether the kitchen executed. Chapter 31 will use it the same way on the P&L.

It fails the moment you use it to compare one dish to another, because it deliberately throws away the one piece of information you need: the size of the plate's price tag. A 20% food cost on a $10 appetizer contributes $8.00. A 35% food cost on a $40 steak contributes $26.00. The percentage says the appetizer is nearly twice as good. The bank account says the steak is three times better.

⚠️ Where the Money Leaks

The food-cost-percentage menu.

Here is what an operator builds when percentage is the only lens they own. Over about eighteen months, item by item, meeting by meeting, always for a defensible-sounding reason:

• The expensive protein comes off, because it "runs 38."
• Portions on the remaining proteins shrink half an ounce at a time.
• The starch-heavy, cheap-input dishes multiply, because they cost 18%.
• Prices drift up on the low-cost items — where a dollar barely moves the percentage — and stay frozen on the high-cost items, where a dollar would.

Twelve months later the food cost report reads 26% and the owner is proud of it. Two things also happened that the report cannot show. The average contribution margin per entrée fell, because the menu shed its biggest dollar earners. And the guests noticed — the menu got smaller, cheaper-looking, and less like a reason to leave the house — so covers fell too.

Fewer covers times a smaller margin is the arithmetic of a slow closure. And the whole way down, the number the operator was watching kept improving. This is the single most common way a competent, attentive, hard-working restaurateur manages themself out of business, and it is exactly the theme this book opened with: the food was never the hard part.

Menu engineering exists to hold both facts at once — what a plate earns, and how often anyone orders it. Which means we need an honest measure of "how often."

12.2 Menu mix: measuring popularity honestly

Menu mix percentage is an item's share of the units sold within its category over a defined period:

$$\text{Menu mix \%} = \frac{\text{units of this item}}{\text{total units in the category}}$$

Three words in that sentence are doing all the work: units, category, and period. Get any one of them wrong and every number downstream is wrong in a way you will not notice.

Units, not dollars. Popularity is a count of choices. If you measure mix in sales dollars, an expensive item looks more popular than it is, and you have smuggled price into an axis that is supposed to measure guest preference. Dollars have their own uses — we will look at three rankings side by side in §12.6 — but the popularity axis counts plates.

Category, not menu. You compare like with like. Dinner entrées against dinner entrées. Appetizers against appetizers. Cocktails against cocktails. If you dump forty items into one matrix, a $12 appetizer that sells 200 times will bury a $31 entrée that sells 54, and the "recommendation" you get back will be to run a restaurant that serves nothing but appetizers. Bellwether's dinner entrée category has exactly five members, which is what a short seasonal menu means (Chapter 10).

Period, stated in advance. A week, four weeks, an accounting period — but chosen before you look at the answer, and held constant when you compare. Picking the period after seeing the data is how people prove whatever they already believed.

Bellwether's week

Here is one full week of dinner service — Tuesday through Saturday, five services — with the Chapter 11 cost cards attached to the Chapter 10 prices.

🧾 Read the Numbers

```text FIGURE 12.2 — "The week the menu told on itself" [the Bellwether plan] THE ARTIFACT Product-mix report, dinner entrées only, one full week (Tuesday through Saturday, five services). Pulled from the point-of-sale Sunday morning; plate costs attached by hand from the Chapter 11 cost cards, which are six weeks old. THE CONTEXT Bellwether, a projected steady-state week in the plan's first year. Five dinner entrées, one wood-fired hearth, one sauté station. 475 covers on the week (95 a night); 330 of them ordered an entrée from this list — a 69.5% entrée attachment.

  ITEM                 UNITS    MIX%     PRICE    COST     FC%      CM    CM TOTAL
  Hearth Chicken          96    29.1%   $29.00   $ 8.52   29.4%  $20.48  $1,966.08
  Bellwether burger       73    22.1%   $21.00   $ 5.10   24.3%  $15.90  $1,160.70
  squash and grains       62    18.8%   $24.00   $ 3.95   16.5%  $20.05  $1,243.10
  pork chop               54    16.4%   $31.00   $10.09   32.5%  $20.91  $1,129.14
  ember trout             45    13.6%   $28.00   $10.40   37.1%  $17.60  $  792.00
  ────────────────────────────────────────────────────────────────────────────────
  TOTALS                 330   100.0%

  entrée sales                                                        $8,739.00
  ideal (theoretical) food cost                                       $2,447.98
  blended food cost          2,447.98 / 8,739.00                          28.0%
  total contribution margin  8,739.00 - 2,447.98                      $6,291.02
  weighted average CM        6,291.02 / 330                              $19.06

WHAT IT SHOWS Five items produce $6,291.02 a week of contribution margin, which annualizes to $327,133. The blended food cost is 28.0% against a 30% plan target — two points to the good. And the five margins fit inside a band of $5.01, with the worst food cost percentage on the menu (trout, 37.1%) out-earning the second-best (burger, 24.3%) by $1.70 a plate. WHAT IT DOESN'T This is IDEAL cost, not actual: it assumes every portion was right, nothing was wasted, comped, dropped, or over-plated (Chapter 11). It is one week, and each share carries roughly two points of sampling noise. It covers $8,739 of the roughly $21,850 of weekly dinner sales — 40.0% — so appetizers, sides, desserts, brunch, and every drop of beverage are outside the frame. It knows nothing about labor, station load, or what a guest would have ordered if an item weren't on the page. THE DECISION Do not act on it yet. Pull three more weeks, re-verify the two cost cards that moved, then place the items and work only the two that sit closest to a boundary — the trout and the pork chop. THE LESSON The product-mix report is the only document in the building that records what your guests actually chose. Attach a current plate cost to it and it becomes the most valuable page you read all month — and it will disagree with your food cost report. ```

Check the mix column yourself, because you will be doing this by hand at some point and the habit is worth building:

• Hearth Chicken: $96 \div 330 = 0.2909 = 29.1\%$
• Bellwether burger: $73 \div 330 = 0.2212 = 22.1\%$
• squash and grains: $62 \div 330 = 0.1879 = 18.8\%$
• pork chop: $54 \div 330 = 0.1636 = 16.4\%$
• ember trout: $45 \div 330 = 0.1364 = 13.6\%$

Sum: $29.1 + 22.1 + 18.8 + 16.4 + 13.6 = 100.0\%$. If your shares do not sum to 100, you have either mis-typed a unit count or you have items in the category that you forgot to include — which is a surprisingly common error when a special ran for three nights.

The popularity index, and where the 70% rule comes from

With five items, if guests chose at random, every item would take 20.0% of the units. That is the expected share, and it gives us a cleaner way to talk about popularity than raw percentages.

The popularity index is an item's actual mix divided by its expected mix:

$$\text{Popularity index} = \frac{\text{item mix \%}}{1 \div \text{number of items}}$$

The indices sum to 5.00, which is a useful arithmetic check: with five items, they always must.

Now the convention. The menu-engineering framework as it is practiced descends from work by Michael Kasavana and Donald Smith in the early 1980s, which took portfolio thinking — the stars-and-dogs vocabulary that was circulating in strategy consulting at the time — and applied it to a restaurant menu. Their popularity threshold, and the one built into essentially every point-of-sale report you will ever see, is 70% of the expected share:

$$\text{Threshold} = 0.70 \times \frac{1}{n} = 0.70 \times \frac{1}{5} = 0.70 \times 20.0\% = \mathbf{14.0\%}$$

An item at or above 14.0% of units is "high popularity." Below it, "low popularity."

Be honest with yourself about what that 70% is. It is a convention, not a finding. It is a reasonable one — it says an item earns its place if it does at least seventy cents on the dollar of an average item's work — but there is no body of evidence establishing 70% as the right number rather than 60% or 80%, and the trade literature that repeats it generally repeats it without support. You should know what it does to your answer before you accept it.

Look at what moved. Under the standard 70% rule the pork chop and the squash are Stars — protect them, never discount them. Set the bar at the average instead and both become Puzzles — items you are told to reposition or reprice because they underperform. The choice of a threshold nobody discusses changes the recommendation for two of five items. When your POS prints a quadrant next to an item, find out what threshold it used. Many of them do not tell you.

Sample size, and the Thursday that lied

Chapter 11 flagged a sample Thursday at Bellwether where the Hearth Chicken took 32% of entrées against the 25% the plan had assumed, and called it noise without showing you why. Here is why.

That Thursday sold 56 entrées, 18 of them chicken: $18 \div 56 = 32.1\%$. The full week, at 330 entrées, says 29.1%. The plan's 25% assumption was simply low. But the honest question is not "which number is right," it is "how much would a number like this move on its own?"

A share measured on $n$ observations wobbles by roughly

$$\sqrt{\frac{p(1-p)}{n}}$$

which for a 29% share on 56 tickets is about 6.1 percentage points. A single Thursday could land anywhere between roughly 17% and 41% with nothing whatsoever changing about the restaurant. Thirty-two percent is not a signal. It is Thursday.

Run the same arithmetic on the full week, $n = 330$, and the wobble drops to about 2 points per share — better, and still large enough to matter enormously when an item sits near a boundary.

FIGURE 12.3 — The same five shares, with the sampling noise drawn in   [the Bellwether plan]

  n = 330 entrées, one week. The band is roughly where each share would land in
  another week with identical underlying demand. The line being tested is 14.0%.

  ITEM                  MEASURED   LIKELY RANGE      DOES THE BAND CROSS 14.0%?
  ──────────────────────────────────────────────────────────────────────────────
  Hearth Chicken          29.1%    24.1% - 34.1%     no  — safely popular
  Bellwether burger       22.1%    17.5% - 26.7%     no  — safely popular
  squash and grains       18.8%    14.5% - 23.1%     no  — but by half a point
  pork chop               16.4%    12.3% - 20.5%     YES — unresolved
  ember trout             13.6%     9.8% - 17.4%     YES — unresolved

  Two of five items cannot be placed on the popularity axis from one week of data.
  One of them is the item the matrix is about to tell you to delete.

This is not a reason to abandon the tool. It is a reason to run it on four weeks rather than one, and to treat anything within about two points of a boundary as unplaced rather than as a recommendation. We will come back to this in §12.8 with an unwelcome finding: for an item sitting a point off the line, even a whole season of data will not settle it, and you will have to decide it on judgment. Knowing that in advance is worth more than the matrix is.

⚠️ Where the Money Leaks

Dirty mix data, and the five ways it happens.

Everything in this chapter runs on the assumption that your point-of-sale unit counts are true. They usually are not, and the failures are boringly consistent:

1. The open-food button. Every restaurant has one — "MISC FOOD," "OPEN ENTREE," "SPECIAL" — and every night some of your entrées get rung through it because a modifier was awkward or the line was moving. Those units vanish from the mix and the sales land nowhere useful.
2. Specials rung as the nearest neighbor. A hearth-roasted duck special goes out as "Hearth Chicken" 41 times because that is the button the server could find. Your Star's popularity is now partly someone else's.
3. Modifiers that are really different dishes. A half portion, a gluten-free build, a substitution that changes the protein — if these ride as modifiers, one item's unit count is quietly averaging two different plate costs.
4. Comps and voids. A comped entrée was made, plated, and eaten; it just was not paid for. Does your report count the unit? Some do, some don't, and some count it and the re-fire. Chapter 34 deals with comp and void discipline properly; for now, find out which convention your report uses before you trust a share.
5. Items renamed or re-priced mid-period. "Winter squash and grains" becomes "spring pea and grains" and the POS creates a second item. Your four-week report now shows two dishes at half the mix each, and both look like Dogs.

Every one of these makes a real item look less popular than it is, which biases the whole exercise toward cutting. Before your first menu-engineering meeting, spend ninety minutes in the POS cleaning the item list. It is the highest-return ninety minutes in this chapter.

12.3 The matrix: constructing it from your own POS data

You now have both axes. Building the grid is eight steps and about twenty minutes once the data is clean.

BUILDING THE MATRIX — the eight steps                        [practitioner procedure]

  1  Choose the CATEGORY and the PERIOD, and write both down before you look.
  2  Pull UNITS SOLD for every item in the category. Not dollars. Units.
  3  Attach the CURRENT plate cost to each item from the cost cards (Ch. 11).
     Not last year's card. If a card is stale, stop and re-cost it.
  4  CM per item        = menu price - plate cost
  5  Total CM           = sum of (units x CM)
  6  Weighted avg CM    = total CM / total units      <- the horizontal axis line
  7  Mix % per item     = item units / total units
     Threshold          = 0.70 / number of items      <- the vertical axis line
  8  PLACE every item against those two lines, and mark anything within two
     points of either line as "unplaced" rather than as a recommendation.

Steps 6 and 7 are the ones people get backward, so be careful: the margin line is a weighted average — total contribution margin divided by total units — not the simple average of the five CM figures. The simple average of $20.48, $15.90, $20.05, $20.91, and $17.60 is $18.99. The weighted average is $19.06. Seven cents, on this menu; on a menu with a wider CM spread and a lopsided mix it can be dollars, and it can move items across the line. Use the weighted one. It is the one that reconciles to actual money.

🧮 Run the Numbers

Both axes, from the top.

The margin axis. Multiply each item's CM by its units and add them up:

Item	Units	CM	Units × CM
Hearth Chicken	96	$20.48 | $1,966.08
Bellwether burger	73	$15.90 | $1,160.70
squash and grains	62	$20.05 | $1,243.10
pork chop	54	$20.91 | $1,129.14
ember trout	45	$17.60 | $792.00
Total	330		$6,291.02

$$\text{Weighted average CM} = \frac{\$6{,}291.02}{330} = \mathbf{\$19.06}$$

Cross-check it the other way, since this is a book about arithmetic that resolves: entrée sales of $8,739.00 minus ideal food cost of $2,447.98 is $6,291.02. Same number, computed from the other end.

The popularity axis. Five items, so the expected share is $1 \div 5 = 20.0\%$, and the threshold is $0.70 \times 20.0\% = \mathbf{14.0\%}$ — which, in units, is $0.140 \times 330 = 46.2$ entrées a week.

Now measure every item's distance from both lines. This is the step almost nobody does, and it is the one that tells you whether to believe the answer.

Item	CM	vs. $19.06	Mix	vs. 14.0%	In units
pork chop	$20.91 | **+$1.85**	16.4%	+2.4 pts	+7.8
Hearth Chicken	$20.48 | **+$1.42**	29.1%	+15.1 pts	+49.8
squash and grains	$20.05 | **+$0.99**	18.8%	+4.8 pts	+15.8
ember trout	$17.60 | **−$1.46**	13.6%	−0.4 pts	−1.2
Bellwether burger	$15.90 | **−$3.16**	22.1%	+8.1 pts	+26.8

Read the last column and feel your confidence drain out of the exercise. The ember trout — the item the matrix is about to label a Dog and tell you to delete — misses the popularity line by 1.2 units a week. Two more trout across five services. One extra fish on Friday and one on Saturday.

Sell 47 instead of 45 and the item's share becomes $47 \div 332 = 14.2\%$, above the line, and its quadrant — and the recommended action — changes completely.

To scale, so you can see how close it is

FIGURE 12.4 — The five dinner entrées, plotted on both axes          [the Bellwether plan]

   $21.00 ┤               ┆     P  $20.91
          │               ┆
   $20.50 ┤               ┆                                     C  $20.48
          │               ┆
   $20.00 ┤               ┆           S  $20.05
          │               ┆
   $19.50 ┤ PUZZLE        ┆   STAR
   $19.06 ┼───────────────┼──────────────────────────────────────────  avg CM $19.06
   $18.50 ┤ DOG           ┆   PLOWHORSE
          │               ┆
   $18.00 ┤               ┆
   $17.50 ┤              T┆  $17.60
          │               ┆
   $17.00 ┤               ┆
          │               ┆
   $16.50 ┤               ┆
          │               ┆
   $16.00 ┤               ┆                   B  $15.90
          │               ┆
          └─────┬─────────┬─────────┬─────────┬─────────┬─────────┬─────
               10%      14.0%      18%       22%       26%       30%
                      menu mix — share of the 330 entrées sold

   C = Hearth Chicken    P = pork chop    S = squash and grains
   T = ember trout       B = Bellwether burger
   ┆ = popularity threshold, 14.0%    ─── = weighted average CM, $19.06
   Schematic; the vertical scale is compressed below $18.00.

Three things are visible here that a table of quadrant labels hides completely.

First, the trout is touching the line. The T and the ┆ are adjacent characters. In a real menu-engineering meeting, somebody will say "the trout is a Dog" as though that were a fact about the trout. It is a fact about 1.2 units.

Second, the vertical spread is tiny. Every marker sits between $15.90 and $20.91. On a menu with a $40 steak and a $12 pasta, the margin axis would be doing real work — it would separate items by ten or fifteen dollars and the quadrants would mean something. Here, three items are within a dollar eighty-five of the line. The margin axis at Bellwether is decided by pocket change.

Third, there is nothing in the upper-left. The Puzzle quadrant is empty. We will make something of that in §12.5 and again in §12.7, because an empty quadrant is not a description of Bellwether's menu — it is a description of a menu with only five items and tightly clustered margins.

👨‍🍳 On the Line

Where these numbers actually come from at 10 a.m. on a Sunday.

The romance of menu engineering is a two-by-two grid on a whiteboard. The reality is a manager, a laptop, and a cup of coffee in a dining room that still smells like last night's hearth.

You pull the product-mix report for Tuesday through Saturday. You export it, which on most systems means a spreadsheet with more columns than you want and item names in a slightly different format than your cost card file uses, so you spend eleven minutes matching "HEARTH CHKN" to "Hearth Chicken." You paste last month's plate costs into a column and then — this is the part that separates a real analysis from a ritual — you check whether any of them moved.

At Bellwether two would have. The trout came in at a different price in week four. The poultry contract renews in March. If you skip that check, you have built a beautiful grid on top of stale costs, and the vertical axis is fiction.

Then you compute six numbers: total units, total CM, weighted average CM, each item's mix, the threshold, and each item's distance from both lines. Twenty minutes.

Do it yourself for the first six months. Not because a manager cannot, but because the twenty minutes is where you learn your own restaurant — that Thursday is a different menu than Saturday, that the burger spikes when it rains, that the trout does most of its business at the bar. None of that is in the summary. All of it is in the doing.

12.4 The four quadrants and the action each one implies

Two lines make four boxes, and each box gets a name and a default action.

• A star is high margin, high popularity. It earns above the menu's average and sells above the popularity threshold. Protect it. Never discount it, never let it run out, never quietly shrink it, and be extremely careful about moving its price.
• A plowhorse is low margin, high popularity. Guests love it; it does not pay you much. Work the cost and the price, gently, and consider what it attaches to.
• A puzzle is high margin, low popularity. It pays well when somebody orders it and nobody does. Sell it before you touch it — position, description, service — and reprice or replace only if selling fails.
• A dog is low margin, low popularity. It neither earns nor moves. Cut it, fix it, or find out what it is really doing for you — and that last option is the one the name discourages, which is the single biggest problem with this vocabulary.

Here is Bellwether's dinner menu placed against the two lines we computed: the weighted average CM of $19.06 and the 14.0% popularity threshold.

FIGURE 12.5 — Bellwether's five dinner entrées, placed                [the Bellwether plan]

                        LOW popularity                 HIGH popularity
                        under 14.0% of units           14.0% of units and over
                  ┌──────────────────────────────┬──────────────────────────────┐
                  │ PUZZLE                       │ STAR                         │
  HIGH margin     │ earns well, sells poorly     │ earns well, sells well       │
  CM at or above  │                              │                              │
  $19.06          │   ( empty this period )      │   pork chop                  │
                  │                              │     $20.91  ·  16.4%         │
                  │   no item on this menu       │   Hearth Chicken             │
                  │   earns above the average    │     $20.48  ·  29.1%         │
                  │   and sells below the line   │   squash and grains          │
                  │                              │     $20.05  ·  18.8%         │
                  ├──────────────────────────────┼──────────────────────────────┤
                  │ DOG                          │ PLOWHORSE                    │
  LOW margin      │ earns poorly, sells poorly   │ earns poorly, sells well     │
  CM below        │                              │                              │
  $19.06          │   ember trout                │   Bellwether burger          │
                  │     $17.60  ·  13.6%         │     $15.90  ·  22.1%         │
                  │                              │                              │
                  │   misses the popularity      │   the second most popular    │
                  │   line by 1.2 units a week   │   item on the menu           │
                  └──────────────────────────────┴──────────────────────────────┘

  Three Stars, one Plowhorse, one Dog, and no Puzzle at all.

Three Stars out of five items is not the result the trade press prepares you for, and it is worth understanding why it happened before you feel good about it.

It happened because the margin line is an average of these five items and nothing else. Three items sit above it and two below, which is roughly what an average does. It is arithmetically impossible for all five to be Stars, and nearly impossible for four to be, no matter how good the menu is. If Bellwether replaced the burger tomorrow with a dish earning $22.00, the average would rise, and one of today's Stars would fall below it and become a Plowhorse without changing at all. Hold that; §12.7 makes it the central objection.

It also happened because the popularity threshold is generous. At 14.0% against an expected 20.0%, an item has to be genuinely unpopular to miss — and with only five items, each one carries a large share by default. On a twenty-item menu the threshold is $0.70 \div 20 = 3.5\%$, and items fall below it constantly.

What "protect the Star" actually means

The instruction for a Star sounds like a non-instruction: leave it alone. It is not. It is four specific, expensive-to-neglect disciplines.

Never run out. The Hearth Chicken sells 96 a week — call it nineteen or twenty a night. If you 86 it at 8:40 on a Saturday, you have not just lost the $20.48. You have handed a guest who came for one specific dish a substitute, and Chapter 23 will show you what that does to the probability of a second visit. Chapter 14's production planning exists largely to protect Stars.

Never discount it. A Star is the item with the least need for a promotion and the most to lose from one. Discounting it trains your best-selling item to be worth less, and it costs you real money per unit at the highest volume on the menu.

Guard the cost card, not the recipe. A Star's contribution margin is exposed to exactly one thing: input prices. Bellwether's chicken is half a 3.5 lb air-chilled bird at $3.20/lb, and the poultry contract renews in March.

🧮 Run the Numbers

What a forty-cent poultry increase does to the Star.

Suppose the March renewal takes air-chilled chicken from $3.20/lb to $3.60/lb — a 12.5% increase, entirely ordinary.

The half bird is 1.75 lb. At $3.20 it was $5.60; at $3.60 it is $6.30. Rebuild the Chapter 11 card: components rise from $8.35 to $9.05, and the 2% waste allowance brings the plate to $9.05 \times 1.02 = **$9.23** (it was $8.52).

• New food cost: $9.23 \div 29.00 = 31.8% (was 29.4%).
• New contribution margin: $29.00 - 9.23 = **$19.77** (was $20.48).
• Cost of doing nothing: 96 units × $0.71 = **$68.16 a week, or $3,544 a year.**

The chicken is still a Star — the menu's weighted average CM also falls, to $18.86, so the item stays above the line. But you have quietly given away three and a half thousand dollars a year, and no report you receive will announce it. It shows up as a food cost report that reads 28.6% instead of 28.0% and gets described in the Monday meeting as "about flat."

This is what cost drift looks like from the inside. It is Chapter 1's killer number two, arriving one contract renewal at a time, and the counter-measure is the calendar entry in §12.8 that says re-cost the top three items in March.

Give it the position it earns. The Star belongs where the eye goes on the page — Chapter 10's menu panel and sweet spot — and in the server's mouth at the table. If your best dollar earner is buried under a heading in eight-point type while a Plowhorse gets a box and a border, you have engineered the menu backward.

🤝 Hospitality

What the Star is doing that the grid cannot record.

Twenty-nine percent of Bellwether's entrée orders are the Hearth Chicken. Say that out loud: nearly a third of everyone who eats dinner in this room eats the same thing.

That is not only a revenue fact, it is an identity fact. The chicken is the dish people describe to a friend on the phone. It is the reason a table of four with one adventurous eater and three cautious ones can all be happy. It is the thing the room is for, and a very large part of what a guest is buying when they choose Bellwether over the eleven other options within a ten-minute walk.

Now notice what the matrix says about it: $20.48 / 29.1% — STAR. Nine characters. Nothing about identity, nothing about the second visit, nothing about the fact that if you took it off the menu you would not lose 96 entrées a week, you would lose tables.

The grid measures what a dish contributes. It has no way to record what a dish means, and on a chef-driven menu the second thing is often larger than the first. Hold both. The number tells you the chicken is your best earner; the room tells you it is your handshake.

12.5 Working a Star, a Plowhorse, a Puzzle, and a Dog — four worked cases

The quadrant is a diagnosis. What follows is treatment, and treatment costs money whether it works or not, so each of these gets worked all the way to a dollar figure.

Case one — the Star: hold, or take a dollar?

The Hearth Chicken at $29.00, CM $20.48, 96 units, 29.1% mix, $102,236 a year in contribution margin. It is the largest single earner on the menu by a wide margin — more than the pork chop and the trout combined. The chef wants to know whether to take it to $30.

The arithmetic of a price increase always has the same shape: how much volume can I lose and still be no worse off?

At $30.00 the CM becomes $21.48. To match today's $1,966.08 a week you need $1{,}966.08 \div 21.48 = 91.5$ units. You currently sell 96. So you can lose 4.5 units a week — 4.7% of chicken volume — and break even. Anything better than that and the dollar is free.

Four-point-seven percent is a small cushion for the most visible price on your most visible dish. A guest who has eaten the chicken six times knows what it costs. Chapter 3 called the menu a brand artifact; the signature item's price is the single number guests actually remember, and moving it is the loudest possible way to raise prices.

The disciplined answer: if you are going to take a dollar, take it somewhere less visible, and take it during a menu change so the whole page moves at once. Which is exactly what §12.8's cadence is for.

Case two — the Plowhorse: the burger, and three real options

The Bellwether burger at $21.00, CM $15.90, 73 units, 22.1% mix. It is $3.16 below the margin line — the widest gap on the menu — and it is the second-most-ordered item in the building. That combination is what a Plowhorse is.

It also produces $60,356 a year of contribution margin, third-most on the menu. Nobody is deleting this item. The question is how to make it pay more without making guests notice.

🧮 Run the Numbers

Three ways to fix a Plowhorse, priced.

Option A — raise the price a dollar, to $22.00.** CM goes to $16.90. Break-even volume: $1{,}160.70 \div 16.90 = 68.7$ units. You sell 73. You can lose 4.3 units a week — 5.9% of burger volume — and be no worse off. If volume holds entirely: $73 \times \$1.00 = \$73.00$ a week = $3,796 a year.**

Option B — re-cost the plate. The burger is $5.10. Suppose a second bid on buns and a larger aioli batch take $0.45 out of it, to $4.65, with no change a guest could detect. CM goes to $16.35. $73 \times \$0.45 = \$32.85$ a week = $1,708 a year, with zero price risk and zero guest risk.

Option C — attach something. The burger's guests are not buying much else. Add a $4.00 side of hearth-charred greens at a $1.10 plate cost — CM $2.90 — and sell it to one burger guest in three: $24 \times \$2.90 = \$69.60$ a week = $3,619 a year, and the burger's own numbers never move. (Side cost illustrative; you would build the card properly per Chapter 11.)

What you would actually do: B first, because it is invisible and free. Then C, because it is additive rather than extractive. Then A at the next menu change, when the whole page reprices and one line does not stand out.

All three together: about $9,100 a year from the item the grid calls a workhorse and most operators leave alone.

Notice what is not on that list: shrinking the patty. Re-portioning a Plowhorse is the classic move and it is nearly always a mistake, because the reason the item is a Plowhorse is that guests order it constantly — which means a great many people know exactly what it is supposed to look like. Chapter 11 covered portion control as a consistency discipline; using it as a margin lever on your most-ordered dish is how a restaurant gets a reputation for getting smaller.

Case three — the Puzzle: the pork chop, one bad month from now

Bellwether has no Puzzle this period, so we will make one honestly rather than inventing an item.

The pork chop is the highest contribution margin on the menu — $20.91 — and the second-least ordered, at 54 units and 16.4% mix. It clears the popularity line by 2.4 points, which §12.2 showed is inside the noise band. The pork chop is one soft month away from being a Puzzle.

Suppose it runs 38 units instead of 54. Total units become 314; total CM becomes $\$6{,}291.02 - (16 \times \$20.91) = \$5{,}956.46$; the weighted average CM becomes $\$5{,}956.46 \div 314 = \$18.97$. The chop's mix is $38 \div 314 = 12.1\%$, below 14.0%. High margin, low popularity: Puzzle.

Now the standard advice is "reposition, rename, or reprice." Price it out and see what it costs.

If you cut the price $2.00, to $29.00, the CM falls to $18.91. To hold the same total contribution you now need $\$1{,}129.14 \div 18.91 = 59.7 \rightarrow 60$ units. You are at 54. The discount has to generate about eleven percent more volume just to break even — and that is eleven percent on an item that was already not selling, driven by a price change most guests will not consciously register. If volume does not move at all, the discount costs $54 \times \$2.00 = \$108$ a week, or $5,616 a year.

If instead you sell it — move it to the panel position the chicken does not need, rewrite the description to say what it actually is, put it in pre-shift for two weeks, and let the servers taste it — and it goes from 54 to 66 units, you gain $12 \times \$20.91 = \$250.92$ a week, or $13,048 a year. No price change. No cost change. Nothing but attention.

That comparison is the entire Puzzle doctrine. A Puzzle is, by definition, an item that already earns more than average. Discounting it attacks the only thing about it that is working. Sell it first, always, and give the selling a fair eight weeks before you conclude it failed.

Case four — the Dog: cutting the trout, priced honestly

The ember trout. CM $17.60, 45 units, 13.6% mix. Below both lines. The grid says Dog, the vocabulary says cut, and this is where a great many operators stop thinking.

Start with what the item does: $792.00 a week, $41,184 a year, of contribution margin. It is the smallest earner on a five-item menu, which is a different sentence from "it does not earn."

Then ask the question the matrix cannot: what do those 45 guests do instead?

🧮 Run the Numbers

The cost of cutting the Dog, under one plausible set of assumptions.

Pull the trout. Assume of its 45 weekly guests:

• 27 (60%) order the Hearth Chicken. You gain $20.48 - 17.60 = \$2.88$ each → **+$77.76**
• 11 (25%) order the Bellwether burger. You lose $17.60 - 15.90 = \$1.70$ each → **−$18.70**
• 7 (15%) go somewhere else, because the trout was why they picked Bellwether. You lose their $17.60 outright — and each of them was dining with someone, who also leaves, taking an average $19.06 entrée with them. $7 \times \$17.60 = \$123.20$ plus $7 \times \$19.06 = \$133.42$ → −$256.62

Net: −$197.56 a week, or −$10,273 a year.

Your food cost report, meanwhile, improves. Removing 45 units at 37.1% pulls the blended entrée food cost from 28.0% down toward 26.5%. The number you were watching gets better while the business gets worse — which is §12.1's failure, arriving with a quadrant label attached to give it authority.

Now change one assumption. Make it 80% chicken, 20% burger, nobody leaves: $36 \times \$2.88 = +\$103.68$, $9 \times \$1.70 = -\$15.30$, net +$88.38 a week, +$4,596 a year. Cutting the trout is now a good idea.

The recommendation reversed on an assumption you have no data for. That is not a reason to despair; it is a reason to know which number your answer is actually resting on. Here it is the defection rate, and you can estimate it: ask your servers how often a table asks whether there is a fish, and count how many trout go out to two-tops where the other guest ordered something else.

If the trout stays, it should not stay unchanged. There are three levers, and they are worth different amounts.

Notice that even the third row does not produce a Star. Fixing the margin makes the trout a Puzzle, not a Star, because margin was never the trout's real problem — 1.2 units a week was. Only selling more of it moves it across the axis that is actually binding. An operator who reaches for the cost card when the popularity line is the one being missed has treated the symptom that was easiest to reach.

Look at the middle row, because it is the counter-intuitive one. Taking $1.30 out of the trout's plate cost is a real improvement — $58.50 a week, $3,042 a year — and it does not change the quadrant, because improving one item also raises the menu average it is being measured against. The item chases a line that moves toward it.

That effect has a clean rule of thumb, and it is worth carrying: moving an item's contribution margin by a dollar closes its gap to the average by about a dollar times (one minus its mix share). A popular item is harder to move across the margin line than an unpopular one, because it is a bigger part of the line. It also means a modest cost increase can cross an item in the other direction: the squash and grains sits $0.99 above the average with an 18.8% share, so a cost increase of $\$0.99 \div (1 - 0.188) = \$1.22$ per plate would flip it from Star to Plowhorse. A dollar twenty-two. One bad grain contract.

🔍 Check Your Understanding

1. An item sells 30 units a week at a $12.00 contribution margin. You raise the price $1.50. How much volume can you lose before the change costs you money?
2. Why is discounting a Puzzle usually the worst of the available moves, in one sentence?
3. The trout's plate cost falls by $1.30 and it stays a Dog. Explain to your chef, without algebra, why the quadrant did not change.

(1: today's contribution is $30 \times 12.00 = \$360$; at $13.50 you need $360 \div 13.50 = 26.7$ units, so you can lose 3.3 units — about 11%. 2: because a Puzzle's only strong feature is its margin, and a discount attacks exactly that. 3: because the line it is measured against is the average of the whole menu, and improving the trout raised the average too — the item got better and so did the bar it had to clear.)

12.6 Weighted contribution margin and the whole-menu view

Everything so far has been about individual items. Step back and look at the menu as one instrument.

Weighted contribution margin is the average number of dollars every entrée sold contributes, weighted by how often each item is actually chosen:

$$\text{Weighted CM} = \frac{\text{total contribution margin}}{\text{total units}} = \frac{\$6{,}291.02}{330} = \$19.06$$

Equivalently, and more usefully for thinking about it, it is each item's CM multiplied by its mix share, summed:

That reconciliation matters. If your weighted average does not equal total CM divided by total units, you have a share that is wrong or an item you left out.

This single number is the bridge between the menu and the rest of the book. Chapter 32 will divide fixed costs by a contribution margin ratio to find a break-even point; Chapter 24 will multiply covers by an average check. Weighted CM is where the menu's contribution to both of those actually comes from. Every dollar you move it is worth 330 dollars a week at Bellwether's volume, or $17,160 a year.

The menu, ranked three ways

Before we ask how to move it, look at how differently the same five items rank depending on which column you sort by.

The pork chop is the second-largest generator of revenue on the menu and the fourth-largest generator of money you keep. The squash and grains is the reverse: fourth by revenue, second by margin. A **$3.95 vegetable plate is the second-most-profitable item Bellwether sells**, ahead of a $31 pork chop and a $28 fish, and it gets there purely on the combination of a 16.5% food cost and a respectable 18.8% mix.

If you have only ever looked at a sales report — and most operators have only ever looked at a sales report, because that is the one the POS emails you every morning — you would have the second and fourth items exactly backward.

🧮 Run the Numbers

Two ways to move the weighted average, compared.

Lever one: change the mix. Suppose brilliant service work moves 20 guests a week off the burger and onto the pork chop — the largest realistic mix shift on this menu, from the worst earner to the best.

Gain: $20 \times (\$20.91 - \$15.90) = \$100.20$. New total CM $6,391.22 on the same 330 units. New weighted CM: $\$6{,}391.22 \div 330 = \mathbf{\$19.37}$ — up 31 cents, worth $10,230 a year.

Lever two: change the prices. Add $1.00 to all five entrées. Every CM rises by exactly $1.00, so the weighted average rises by exactly $1.00, to **$20.06** — worth $17,160 a year, if volume holds.

The price lever is more than three times the mix lever, and it took an afternoon.

Here is why, and it is the most important structural fact about this particular menu. The theoretical ceiling for weighted CM — if every single guest ordered the highest-margin item — is $20.91. The floor, if everyone ordered the burger, is $15.90. **The entire range that mix can move is $5.01**, and Bellwether already sits at $19.06, which is 63% of the way up it. There is $1.85 of headroom above and you will never capture more than a fraction of it.

The tighter your contribution margins cluster, the less menu engineering can do through mix, and the more your leverage sits in pricing and costing. On a menu with a $12 pasta and a $46 steak, the reverse is true and mix work is enormously powerful.

That conclusion deserves to be stated plainly, because it is the payoff of the observation we made back in §12.1: four of Bellwether's five items sit within $3.31 of each other on contribution margin, and all five within $5.01. That flatness is not an accident. It is what happens when a chef prices a short menu by feel and a decent instinct — every plate ends up worth roughly the same to the house, which is a defensible thing for a menu to be. It also means the two-by-two grid, run on this menu, is going to spend most of its time arguing about pocket change while the real levers sit in the purchasing office and on the price line.

Know which kind of menu you have before you decide how much to invest in engineering it.

⚖️ Code and Compliance

What you may and may not say when you change a menu.

Re-engineering means reprinting, and reprinting means claims. A few things to have straight before the file goes to the printer — every one of which varies by state, county, and city, so verify locally and, for anything consequential, with an attorney:

• The printed price must be the price charged. Consumer-protection and weights-and-measures enforcement in most jurisdictions treats a POS price that exceeds the menu price as a straightforward violation, not a clerical matter. When you reprice, the menu and the POS change the same night. Chapter 34 makes this a control, not a good intention.
• Descriptive claims are enforceable claims. "Local," "wild-caught," "grass-fed," "organic," "house-made," and a named farm are representations about the product. Fish is a particular flashpoint: species substitution and wild-versus-farmed claims have been the subject of repeated public enforcement and journalism. If you call the trout wild, it had better be wild, and you had better be able to produce the invoice.
• Allergen accuracy survives a re-cost. The most dangerous moment for an allergen is a quiet ingredient substitution made for cost reasons. If §12.5's re-costing swaps an oil, a stock, or a thickener, the allergen matrix and the server training change with it. Chapter 25 has the protocol.
• Calorie disclosure under federal menu-labeling rules attaches to chains at or above a threshold number of locations. A single independent like Bellwether is generally outside it, but some state and local rules reach further — check before you assume you are exempt.

None of this is a reason to reprint less often. It is a reason to keep a checklist in the menu file.

12.7 The limits of the matrix: what it ignores and how it misleads

This is the most important section in the chapter, and if you take one thing from it, take this: the menu-engineering matrix is a description of two variables out of about nine that matter. It is genuinely useful. It is also, run carelessly, an authoritative-sounding machine for destroying good menu items, and the four-animal vocabulary makes it worse by attaching a verdict to a measurement.

Here is everything it cannot see.

It has no labor axis at all

The matrix costs ingredients. It does not cost the hands.

Two of Bellwether's Stars are not remotely equivalent on a Saturday at 7:45. The Hearth Chicken goes into the wood-fired hearth in advance, comes out, gets finished with salsa verde, and leaves the pass in well under a minute of a cook's attention. The squash and grains is à la minute on the sauté station — grains warmed, squash seared, sauce built, plated to order.

Suppose — and this is illustrative, since real capacity comes from Chapter 14's ticket-time work — that the hearth can deliver 40 chickens an hour and the sauté station can deliver 22 squash plates an hour.

On the matrix, these two items are 43 cents apart. On a Saturday night when the sauté station is the bottleneck, one of them is worth nearly twice the other per unit of the thing you actually run out of. And it gets worse: the squash plate ties up the station that also makes three appetizers, so its real cost includes the appetizers that did not go out.

Contribution margin per plate is the right number when your constraint is guests. Contribution margin per minute of the constrained station is the right number when your constraint is the kitchen. Most full-service restaurants are constrained by guests on Tuesday and by the kitchen on Saturday, which means the correct menu is different on different nights — and the matrix has no way to say so. Chapter 14 handles station load and ticket times; Chapter 19 puts a dollar on the labor hour.

It treats items as independent, and they are not

This is the failure with the largest dollar consequences, and it has a name from Chapter 10: cross-utilization. One purchase order becomes several menu lines. Cut one line and you have not removed a cost — you have removed a use, and everything that shared the input gets more expensive.

FIGURE 12.6 — What one whole-bird program actually feeds            [the Bellwether plan]

   ONE STANDING ORDER OF AIR-CHILLED CHICKEN  (whole birds, $3.20/lb)
        │
        ├── halves ────────▶  HEARTH CHICKEN, the $29 dinner entrée      ← the Star
        │
        ├── backs, necks ──▶  house stock ──┬─▶ pan sauce, Hearth Chicken
        │                                   ├─▶ pan sauce, pork chop
        │                                   └─▶ weekend brunch soup
        │
        ├── livers ────────▶  chicken-liver mousse, the bar snack list
        │
        ├── wings ─────────▶  Thursday bar special
        │
        ├── rendered fat ──▶  schmaltz ─▶ the roasted roots on two plates
        │
        └── trim ──────────▶  family meal   (not sold; a labor and morale line)

   Six sellable lines and one staff meal come out of one purchase order. The
   matrix rates line 1. It cannot see lines 2 through 6, and it has no idea
   that killing line 1 would force you to BUY stock, livers, wings, and fat.

Bellwether's chicken is a Star, so nobody is cutting it — but run this thought through your own menu and notice how quickly the independence assumption falls apart. The pork chop's pan sauce depends on a poultry program that exists because of a different dish. Delete the chicken and the pork chop's plate cost rises, which lowers its contribution margin, which moves it on the matrix. One deletion, two items changed, and the grid you are staring at knows about neither.

The same logic runs the other way on the trout. A twice-weekly seafood delivery exists because there is enough seafood on the menu to justify it. Take the fish off and whatever else rides on that delivery — mussels, oysters, the smoked-fish snack at the bar — is suddenly carrying the freight and the order minimum alone. Delivery minimums and freight thresholds are the most common way one item's volume quietly subsidizes another item's price, and Chapter 13 will make you look at them properly.

⚠️ Where the Money Leaks

The deletion that raised food cost.

The pattern, which you will see if you operate long enough: an operator runs the matrix, cuts three Dogs at once, and reprints. Two months later food cost is up a point and a half and nobody can explain it.

What happened is that the three deleted items shared inputs with the survivors. The mushroom that appeared on two plates now appears on one, so the case does not turn over and half of it goes to waste. The herb that justified a small-farm delivery is now a single-use item bought from the broadline at a worse price. The braise that used the trim from the item you cut is now built from a primal you buy specifically for it.

On a menu built with real cross-utilization, cutting items raises the unit cost of the items you keep. How much depends entirely on your specific menu, which is why nobody can give you a rule — and why the disciplined move is to cut one item at a time and re-cost the survivors afterward.

It counts plates, not people

Bellwether serves 475 covers on the week and sells 330 entrées from this list. That is a 69.5% attachment: nearly a third of the guests in the room are not represented in the matrix at all. They came for the bar, they split something, they ate three appetizers, they came to brunch.

Worse, the matrix cannot see veto power. A party of four picks a restaurant by consensus, and consensus is usually driven by the most constrained eater at the table. The one guest who wants fish, or the one who wants a vegetable main, decides where four people go. The trout gets credited with one unit; it may have been responsible for four covers and $184 of check.

This is a real and countable thing, and it is the reason the grid's advice on a Dog is so dangerous. Chapter 23 will work the arithmetic of the return visit. For now hold the shape of it: an item's value to the business is its contribution margin plus its contribution to the decision to come at all, and only the first term is on the page.

The axes are relative, so every menu produces Stars and Dogs by construction

Here is the objection that ought to change how you use this tool forever.

Both lines are computed from your own menu. The margin line is your weighted average. The popularity threshold is 70% of your expected share. Nothing about either line refers to the outside world, a benchmark, or a standard of good. Put five outstanding items on a menu and the matrix will still find you a Dog. Take the Dog off and it will find you another one.

Watch it happen with Bellwether's actual numbers. Assume, for the sake of the demonstration, that nobody substitutes — which is exactly the assumption the matrix implicitly makes.

FIGURE 12.7 — What happens when you keep taking the advice           [the Bellwether plan]

  ROUND 1 — five items            weighted avg CM $19.06    threshold 14.0%
    pork chop          $20.91 / 16.4%    STAR
    Hearth Chicken     $20.48 / 29.1%    STAR
    squash and grains  $20.05 / 18.8%    STAR
    ember trout        $17.60 / 13.6%    DOG        ◀ the matrix says cut it
    Bellwether burger  $15.90 / 22.1%    PLOWHORSE

  ROUND 2 — cut the trout, four items, nobody substitutes
                                  weighted avg CM $19.29    threshold 17.5%
    pork chop          $20.91 / 18.9%    STAR
    Hearth Chicken     $20.48 / 33.7%    STAR
    squash and grains  $20.05 / 21.8%    STAR
    Bellwether burger  $15.90 / 25.6%    PLOWHORSE  ◀ now $3.39 below the line
                                                      (it was $3.16 below)

  ROUND 3 — cut the burger too, three items left
                                  weighted avg CM $20.46    threshold 23.3%
    pork chop          $20.91 / 25.5%    STAR
    Hearth Chicken     $20.48 / 45.3%    STAR       ◀ clears the line by 1.6 cents
    squash and grains  $20.05 / 29.2%    PLOWHORSE  ◀ a STAR two rounds ago

  Nothing about the squash and grains changed. Not the recipe, not the price,
  not the plate cost, not one guest's opinion of it. The average moved.

Follow that carefully. In Round 3 the Hearth Chicken — Bellwether's signature dish, the anchor of the concept, the largest earner in the building — clears the margin line by less than two cents. One more round and the book's own hero dish is a Plowhorse.

The mechanism is simple and inescapable. Every time you delete the worst item, the average of what remains goes up, which means the bar rises for everyone still standing. The matrix is a ranking wearing the costume of a standard. Its advice, followed literally and repeatedly, terminates at a one-item menu.

The correct use is the opposite of literal: treat the quadrants as a way of generating questions in priority order, never as a verdict. "The trout is below both lines — why, and what is it doing for us?" is a good question. "The trout is a Dog" is a conclusion the arithmetic did not earn.

It is one period, and it is the past

Three more limits, briefly, each of which has bitten operators badly.

Sample size. Covered in §12.2, and worth repeating because it is the one people wave away: on a single week, two of Bellwether's five items cannot be placed at all. The band around a 13.6% share on 330 entrées runs from roughly 9.8% to 17.4%, and the line being tested sits inside it.

Seasonality. Bellwether runs a short seasonal menu. The squash-and-grains dish does not exist in June — by design. Comparing an October mix to a June mix compares two different restaurants. And even the surviving items move: the burger's share rises on the patio in summer, the chop's rises in February. Run the matrix within a menu, not across menu changes, and never compare a holiday period to an ordinary one.

It is entirely backward-looking. The matrix reports what guests chose from the page you gave them. It has nothing to say about an item you have not launched, the effect of a price you have not tried, a competitor who opened last month, or a dish that is fading because it has been on for two years. It is a rearview mirror with excellent resolution.

It has no beverage in it, and beverage is where the margin lives

The last limit is the one with the largest number attached, and it can flatly reverse the chapter's own example.

Bellwether's plan is 28% beverage, and Chapter 15 will show you why: pour cost around 22% against food cost around 30%. An entrée that pulls a drink is worth more than an entrée that does not, and the matrix cannot see a drop of it.

🧮 Run the Numbers

The trout, with the wine attached.

Assume — illustratively, since real attachment comes from your own POS — that a trout guest orders a $14.00 glass of white at roughly a 22% pour cost, and a burger guest orders a $7.00 beer at roughly 24%.

	Ember trout	Bellwether burger
Entrée CM	$17.60 | $15.90
Drink	$14.00 glass | $7.00 beer
Drink cost	$3.08 | $1.68
Drink CM	$10.92 | $5.32
Total contribution per guest	$28.52** | **$21.22

The Dog now out-earns the Plowhorse by **$7.30 a guest** — more than four times the $1.70 gap the food-only analysis showed.

Run it across the trout's 45 covers a week: $45 \times \$7.30 = \$328.50$ a week, or $17,082 a year, of advantage that the entrée matrix simply does not contain.

This does not prove the trout should stay. The attachment rates are assumed, and you would have to measure them. It proves that a two-variable model can be reversed by a third variable you left out — and that the third variable here is 28% of the plan's revenue and the best margin in the building. Chapter 15 will teach you to pull attachment data properly; Chapter 16 will do it for wine.

The honest summary

Nine things the matrix cannot see: labor, station load and throughput, cross-utilization, guest counts as distinct from item counts, veto power and the decision to visit at all, substitution, seasonality, sample size, and beverage. Plus one structural flaw: both axes are relative to the menu you already have.

Use it anyway. It is still the fastest way to turn a product-mix report into a short, ranked list of questions, and a short ranked list of questions is worth a great deal on a Monday morning. Just never let a four-letter animal make a decision that a paragraph of thinking would have made better.

🔍 Check Your Understanding

1. Two Stars have identical contribution margins. One takes 40 seconds of a cook's attention, the other four minutes on the busiest station. Are they equally good items? On which nights?
2. Explain, in one sentence, why cutting the item at the bottom of the matrix guarantees that a new item will appear at the bottom of the matrix.
3. Bellwether's entrée matrix covers $8,739 of roughly $21,850 in weekly dinner sales. What share is that, and name three things in the remainder that could change a quadrant's recommendation.

(1: No. They are equivalent on a Tuesday when guests are the constraint and very different on a Saturday when the station is; the second item's true cost includes whatever else that station could not produce. 2: Because the lines are averages of the surviving items, so removing the worst one raises the average and pushes someone else below it. 3: 40.0%; the remainder includes appetizers, sides, desserts, brunch, and all beverage — beverage attachment, appetizer attachment, and shared inputs with off-matrix items could each reverse a recommendation.)

12.8 The re-engineering cycle: how often, and what to change first

Re-engineering actions are the moves available once an item is placed: sell it, re-cost it, re-price it, re-portion it, reposition it on the page, or replace it. They are not equal. They differ in cost, in visibility to the guest, in reversibility, and in how long they take to show a result — and almost every operator reaches for them in exactly the wrong order.

The cadence

Three different clocks, running at three different speeds.

That last line contains a genuinely annoying constraint that nobody mentions. If you change the menu four times a year, no item cohort ever accumulates more than ten to twelve weeks of clean data before the page changes underneath it. At Bellwether that is roughly 3,300 to 3,960 entrées.

Run the noise arithmetic on a full season at $n = 3{,}300$ and the band around the trout's 13.6% share narrows to about 12.4% to 14.8%. The 14.0% line is still inside it.

Sit with that. An item within a point of the popularity threshold will never be resolved by data. Not in a week, not in a month, not in a whole season. If you are waiting for the numbers to tell you what to do about the trout, you will wait until the menu changes and the question disappears unanswered.

This is not a failure of your systems. It is the ordinary condition of running a small restaurant, and the right response is the practitioner's one: gather what data you can, notice when it has stopped being able to answer the question, and then decide on judgment — deliberately, on the record, with the reasoning written down. The written reasoning is what lets you evaluate the decision later, which is the only thing that turns judgment into experience.

The order of operations

FIGURE 12.8 — What to change first                            [practitioner convention]

  cheapest, least visible ──────────────────────▶ most expensive, most visible

  1  FIX THE DATA        Current cost cards. Clean POS buttons. One category,
                         one period, stated in advance. Anything you do before
                         this step is guessing more loudly.

  2  SELL IT             Position on the page, the description, pre-shift, the
                         server who names it at the table. Costs nothing,
                         reverses instantly, and moved the pork chop $13,048 a
                         year in §12.5. Try it first. Always.

  3  RE-COST IT          A yield test, a spec change, a bigger batch, a second
                         bid. The guest never sees it. Raises CM without
                         touching the price. Chapter 11 and Chapter 13.

  4  RE-PRICE IT         Visible, reversible, and the fastest lever you own.
                         One dollar across all five entrées is $17,160 a year
                         at this volume. Do it at a menu change, not alone.

  5  RE-PORTION IT       Visible in a way guests remember and repeat. Only
                         after 2, 3, and 4, and only where the portion was
                         genuinely wrong rather than merely expensive.

  6  REPLACE OR CUT      Costs you the item, the guests who came for it, and
                         whatever else its purchase order was subsidizing.
                         One at a time, and re-cost the survivors afterward.

  Most operators start at 6. It is the most expensive possible order, and it is
  the one the matrix's vocabulary actively encourages — because "Dog" sounds
  like an instruction and "reposition on the page" does not.

Change few things, and change them one at a time

The discipline that separates a re-engineering program from a re-engineering event is attribution.

If you reprice three items, re-cost two, redesign the menu layout, and put a new special on — all at a seasonal change, which is when the temptation is strongest — and the weighted contribution margin moves forty cents, you have learned nothing. You do not know which move produced it, whether one of them actually cost you money, or whether the whole thing was weather.

The rule that works: two or three deliberate changes per cycle, each one written down with the number you expect it to produce, and a review at the next cycle against that number. It feels frustratingly slow. Over two years it produces a restaurant whose operator knows, from evidence, what their guests respond to — which is a durable asset that no consultant can sell you.

👨‍🍳 On the Line

Running the meeting.

The menu-engineering meeting is 45 minutes, once a period, and it has exactly three people in it: the chef, the person who runs the floor, and whoever owns the numbers. At Bellwether that is both partners and nobody else.

The agenda, in order:

1. Are the cost cards current? If two are stale, the meeting is now a costing meeting. Reschedule the rest. This happens more often than anyone admits.
2. What moved since last period, and by how much? Not quadrants — movement. An item whose share dropped four points is more interesting than an item that has been a Dog for six months.
3. Which items are within two points of a line? Those are unplaced. Name them out loud as unplaced, so that nobody in the room argues about a verdict the data did not produce.
4. Two changes. What are they, what do we expect each one to be worth, and who owns it?
5. Read back last period's two changes against what we expected.

What kills this meeting: the chef defending a dish as though the numbers were an insult, and the numbers person treating a dish as a row. Both are avoided the same way — by agreeing in advance that the matrix generates questions, not verdicts, and that the person who cooks the food gets the first answer to every question about it.

The floor person is not decoration here. They are the only one who knows why guests hesitate at the trout and which item gets asked about by people who then order something else. That is data too. It does not come out of a POS.

🍽️ The Business Plan

Checkpoint 12 of 40 — the menu-engineering appendix.

Chapter 10 built Bellwether's draft menu. Chapter 11 costed it. This chapter turns those two documents into the appendix a plan reader actually wants: not a list of dishes and not a list of costs, but evidence that the people writing this plan understand which items carry the business and what they intend to do about the rest.

🧾 Read the Numbers

```text FIGURE 12.9 — "Appendix: Projected Menu Engineering" [the Bellwether plan] THE ARTIFACT One page of the business plan: projected dinner-entrée mix, the contribution-margin ranking, the four-quadrant placement, and the stated first-year actions. THE CONTEXT Pre-opening. There is no POS data because there is no restaurant. The mix below is a FORECAST, built from the concept work in Chapter 2, the menu design in Chapter 10, and the partners' combined experience in comparable rooms.

  PROJECTED WEEK — 475 covers, 330 dinner entrées, 5 services

  ITEM                 UNITS   MIX%       CM   WEEKLY CM   QUADRANT
  Hearth Chicken          96   29.1%  $20.48   $1,966.08   STAR
  Bellwether burger       73   22.1%  $15.90   $1,160.70   PLOWHORSE
  squash and grains       62   18.8%  $20.05   $1,243.10   STAR
  pork chop               54   16.4%  $20.91   $1,129.14   STAR
  ember trout             45   13.6%  $17.60   $  792.00   DOG
  ─────────────────────────────────────────────────────────────────
  TOTALS                 330  100.0%           $6,291.02

  weighted average CM          6,291.02 / 330          $19.06
  popularity threshold         0.70 x (1/5)             14.0%
  entrée sales, weekly                              $8,739.00
  ideal entrée food cost                            $2,447.98  = 28.0%
  annualized entrée contribution margin           $327,133.04

WHAT IT SHOWS Five entrées carry $327,133 of annual contribution margin, on $454,428 of annual entrée sales — 29.3% of the $1,550,000 revenue plan. The projected mix produces a 28.0% blended entrée food cost against the plan's 30% food target. The menu is not dependent on any single dish for margin: the top four items are within $3.31 of each other on CM, and the largest earner is a Star at a defensible 29.4% food cost. WHAT IT DOESN'T It does not settle the mix, because the mix is the one number in this appendix with no evidence behind it. It is ideal cost, not actual (Chapter 11). It covers 40.0% of dinner sales and none of brunch or beverage. It contains no labor, so a Star that ties up the sauté station reads identically to one that does not. And two of the five placements — the trout and the pork chop — sit inside the sampling noise of any single period, so they are projections of a coin's edge. THE DECISION Present the 28.0% as a target the menu is CAPABLE of, not as a two-point saving to be spent. Commit in writing to running the matrix at the end of period one, four, and eight, and to naming the trout decision at period four rather than letting it drift. THE LESSON A projected mix is a forecast wearing a spreadsheet's clothes. The honest way to present one is with the arithmetic shown, the fragile numbers flagged, and a date attached to the moment you will replace it with real data. ```

What this chapter adds to the plan. The menu-engineering appendix above, plus three commitments written into the operating section:

What this checkpoint does not settle, stated plainly because the plan is stronger for saying it:

• The mix is a guess. Every other number on the page is arithmetic; this one is judgment. Week four of actual service will teach the partners more about their mix than the entire planning process did.
• The two points are not banked. A 28.0% ideal entrée food cost against a 30% target is worth roughly $22,320 a year on $1,116,000 of food sales — if actual matches ideal, which it never fully does, and if the mix holds, which it may not. The plan takes the 30% target and treats 28.0% as headroom for the waste, comps, and over-portioning that Chapter 11 warned are coming.
• Nothing here contains labor. The squash and grains is the second-largest margin generator on the menu at a $3.95 plate cost, and it is also the most labor-intensive plate on the line. The plan cannot resolve that tension until Chapter 14 puts station capacity on it and Chapter 19 puts a dollar on the hour.

Open questions carried forward:

1. Does the projected mix survive four weeks of real guests — and what does the room do to it? (Chapters 22, 24)
2. Is the ember trout a Dog, a Plowhorse, or the reason a table chose Bellwether at all? (Chapters 13, 23)
3. What does the March poultry renewal do to the Star's cost card, and does the price move with it? (Chapter 13)
4. Does the squash-and-grains plate — a $3.95 vegetable dish that is the menu's second-biggest earner — survive the June seasonal change, and what replaces it? (Chapter 10, revisited)
5. What does this matrix look like when the vertical axis is contribution margin per minute of constrained station time rather than per plate? (Chapters 14, 19)

Conclusion

A menu is a portfolio, and this chapter gave you the two numbers that describe every position in it: what a plate earns and how often anyone orders it.

The earning number is contribution margin — price minus plate cost, in dollars — and it is the axis because dollars are what pay rent. Bellwether's own menu makes the case better than an argument could: the item with the worst food cost percentage on the page, the trout at 37.1%, out-earns the item with the second-best, the burger at 24.3%, by $1.70 on every plate. An operator running the menu on percentage would cut the wrong dish and congratulate themselves on the report.

The frequency number is menu mix percentage, and it is honest only when you count units rather than dollars, compare inside a category rather than across the whole menu, and choose your period before you look at the answer. Bellwether's five entrées come in at 29.1%, 22.1%, 18.8%, 16.4%, and 13.6% — which sums to 100.0%, as it always must — against a conventional 70% threshold of 14.0%. Three Stars, one Plowhorse, one Dog, no Puzzle. The weighted contribution margin is $19.06 a plate, $6,291.02 a week, $327,133 a year.

And then §12.7 spent a long time taking the tool apart, because a book that sold you a formula without its failure modes would have failed you. The matrix has no labor axis, no way to see cross-utilization, no beverage in it, no guest counts, no substitution, no seasonality, and not nearly enough data — the trout misses the popularity line by 1.2 units a week, which is inside the noise of a full season's sample. Worst of all, both axes are relative to your own menu, so the tool will keep finding you a Dog until you have one item left. Figure 12.7 showed the Hearth Chicken itself clearing the margin line by under two cents after two rounds of taking the advice literally.

Use it to generate questions in priority order. Never let it hand down a verdict.

What remains open is everything the matrix left out, and the next two chapters are where it comes back. Chapter 13 goes to the loading dock, because contribution margin is decided by what you paid and what you actually received — specs, par levels, receiving, the walk-in, and the count sheet that turns ideal food cost into real food cost. Chapter 14 goes to the line, where a plate's cost in minutes finally gets a number, and where you will find out that two Stars can be very different items on a Saturday at eight o'clock.

Key Terms

Menu engineering — analyzing a menu as a portfolio by plotting each item's contribution margin against its share of units sold, classifying items into four quadrants, and taking a different action in each. The framework in common use descends from work by Kasavana and Smith in the early 1980s. (Ch. 12)

Contribution margin (CM) — menu price minus plate cost: the dollars a dish contributes toward labor, occupancy, every other cost, and profit. The profit axis of the matrix, and the number an operator banks. (Ch. 12)

Menu mix percentage — an item's units sold divided by total units sold in its category over a stated period. Counts plates, not dollars; compared within a category, never across the whole menu. (Ch. 12)

Popularity index — an item's menu mix divided by its expected share ($1 \div n$). An index of 1.00 means the item sells exactly as often as an average item; the conventional threshold for "popular" is 0.70. (Ch. 12)

Star — high contribution margin, high popularity. Protect it: never discount it, never let it run out, guard its cost card, and give it the best position on the page. (Ch. 12)

Plowhorse — low contribution margin, high popularity. Guests love it and it pays you little. Re-cost it first, attach something to it, and reprice it at a menu change. (Ch. 12)

Puzzle — high contribution margin, low popularity. It pays well and nobody orders it. Sell it — position, description, service — before you consider repricing, because a discount attacks the only thing about it that works. (Ch. 12)

Dog — low contribution margin, low popularity. The quadrant that says "cut it" and the one that most deserves a second question: what is this item doing for the guests who order it, for the items that share its inputs, and for the decision to come at all? (Ch. 12)

Weighted contribution margin — total contribution margin divided by total units sold; equivalently, the sum of each item's CM multiplied by its mix share. It is the horizontal line of the matrix and the menu's single summary number. Bellwether's is $19.06. (Ch. 12)

Re-engineering actions — the six moves available once an item is placed: sell it, re-cost it, re-price it, re-portion it, reposition it, or replace it — ordered here from cheapest and least visible to most expensive and most visible. Most operators start at the wrong end. (Ch. 12)

Spaced Review

1. An item sells 88 units a week at a $14.25 contribution margin; the menu's weighted average CM is $16.40 and the popularity threshold is 11.0% on 640 total units. Place the item in a quadrant, showing both comparisons, and state the first action you would take.
2. Without looking back: why does improving one item's contribution margin sometimes fail to move it into a different quadrant?
3. From Chapter 11: Bellwether's Hearth Chicken costs $8.52 and sells for $29.00. Compute its food cost percentage and its contribution margin, then explain to a skeptical chef why the second number is the one that belongs on a matrix.
4. From Chapter 1: an operator cuts three high-food-cost items, and the following quarter food cost falls two points while operating profit falls as well. Name the two mechanisms from this chapter that would produce that result, and name the failure mechanism from Chapter 1 it resembles.
5. The recurring question: you are handed a product-mix report showing that your highest-margin entrée sells 9% of units against a 14% threshold, and your software has labeled it a Puzzle. Before you change anything, list the four questions you would ask about the data itself — and say what each one could reveal that would change the label.