Case Study 2 — The Menu That Engineered Itself to Death

A labeled composite. This case is constructed from patterns that recur across independent full-service restaurants: the food-cost-percentage spiral, iterative menu cutting, and the cross-utilization collapse. No real restaurant's financials are represented. Every figure is illustrative and internally consistent, and the arithmetic is included so you can check it. It is presented here because a fully public independent-restaurant case with item-level cost and mix data does not exist — those numbers are never disclosed — and a labeled composite that computes teaches the mechanism better than a real case with the numbers missing.

Background

A twelve-year-old neighborhood bistro, 74 seats, dinner six nights, in a second-ring suburb of a mid-size city. Owner-operator, chef-driven, a well-liked room with a long list of regulars. Annual sales around $1.35 million, food 76% / beverage 24%.

The menu at the start of the story: eighteen dinner entrées, which is a large menu for an independent and a fact the owner was privately proud of. Broad appeal, a lot of regulars with a "usual," and a kitchen that had built genuine cross-utilization over a decade — a whole-animal pork program feeding five dishes, a fish delivery three times a week that supported four, a stock program that touched almost everything.

Food cost was running 32.5% against a 30% target. The owner's accountant flagged it. The owner bought a menu-engineering module from their POS vendor, ran it, and got a report with quadrant labels on it.

The operating issue

The report identified six Dogs out of eighteen entrées. That is not surprising and should not have been alarming: on an eighteen-item menu the popularity threshold is $0.70 \div 18 = 3.9\%$, and a menu that broad will always have a tail. The report did not say that. It said "Dog," six times, next to six dishes.

The owner cut all six at the next menu print. The reasoning was sound-sounding at every step: they were below both lines, they were the six highest food-cost items, the report was unambiguous, and cutting them would simplify a kitchen that was undeniably too complex.

Here is what actually happened over the following three periods.

Period 1 — the numbers improve

Food cost fell from 32.5% to 30.1%. Exactly as predicted, because six of the highest-cost items were gone. The owner was pleased and told the accountant so.

Covers were down 4%, which was attributed to weather.

Period 2 — the survivors get more expensive

Food cost came back to 31.4%, and nobody could explain it. The cost cards had not changed. Prices had not changed. Portions were being checked.

What had changed was purchasing. Three of the six deleted items were the ones justifying the three-times-a-week fish delivery. With four fish dishes down to two, the order no longer cleared the distributor's minimum, so the restaurant went to twice weekly at a higher per-pound price and worse freshness on day two. Two more of the deleted items had used the trim from the whole-pork program; with those gone, the program's yield economics stopped working, and the kitchen switched to buying portioned loin at roughly 40% more per usable pound.

Cutting six items raised the plate cost of at least five items that were kept. None of that appeared anywhere in the menu-engineering report, because the report treats items as independent.

Period 3 — the guests, and the second matrix

Covers were down 11% against the same period the prior year. The owner ran the matrix again on the remaining twelve items.

It identified four new Dogs.

This is the moment the case turns on, and it is Figure 12.7 happening in a real building. The four new Dogs were not new problems. They were the bottom of a shorter list, measured against an average that had risen because the previous bottom had been removed. Two of the four had been comfortable Plowhorses three periods earlier and had not changed in any respect.

The owner, to their credit, stopped. But by then the restaurant had lost the two things that were hardest to get back: a menu broad enough that a party of six could always find something, and a purchasing structure built over a decade that made a 30% food cost achievable at all.

What it shows

The report was not wrong. It was answering a different question than the one the owner asked. The matrix correctly identified which items were below the menu's own two averages. The owner heard "which items should I cut," which the matrix has no capacity to answer, because answering it requires knowing about substitution, cross-utilization, veto power, and the guest's reason for coming — none of which is in the input data.

Iterative cutting is a ratchet. Every deletion raises the average and the threshold, which manufactures a new bottom quadrant. An operator who "follows the data" repeatedly will keep receiving cut recommendations until the menu is very short, and each recommendation will look as well-founded as the last.

Cross-utilization is invisible until it breaks. The fish delivery and the whole-pork program were not line items on any report. They were operating structures, and their economics depended on volume across several dishes. The failure showed up as an unexplained food-cost increase two periods after the cause, which is exactly the lag Chapter 1 described for cost drift — long enough that nobody connects it to the decision.

A high food cost is a symptom, and the diagnosis was never made. The original problem was 32.5% against a 30% target — 2.5 points on roughly $1.03 million of food sales, about $25,700 a year. That gap could have been portioning, waste, purchasing, uncosted specials, comps, menu drift, or simply prices set three years earlier against costs that had moved. Chapter 11's ideal-versus-actual comparison would have separated those in a week. The owner reached for the menu instead of the variance, and cut items to fix a problem that may not have been in the items at all.

Outcome

Two years later the restaurant is open, doing roughly $1.18 million, with a fourteen-item menu, a food cost around 30.5%, and a covers number that never fully recovered. It is a viable business. It is a meaningfully smaller one, and the owner's own account is that the menu decision cost them a tier of regulars who "stopped finding what they came for."

Nothing catastrophic happened. That is the point. This is Chapter 1's bleeding pattern, not its lightning pattern — a series of individually defensible decisions, each supported by a number, adding up over eighteen months to a business that is a hundred and seventy thousand dollars smaller.

The lesson

Menu engineering generates questions. It does not generate verdicts, and a report that prints a verdict is more dangerous than one that prints nothing — because it supplies the confidence to act without supplying the information required to act well.

Three specific disciplines would have prevented every stage of this:

1. Diagnose before you cut. A food-cost gap is a variance problem until proven otherwise. Run ideal versus actual first.
2. One item at a time, and re-cost the survivors. Six simultaneous deletions made attribution impossible and hid a purchasing collapse for two full periods.
3. Ask what each item is doing that the report cannot see — who orders it, what it shares an invoice with, and whether it is the reason a table chose you.

Discussion questions

1. Compute the popularity threshold for an eighteen-item menu and a twelve-item menu. Explain, using those two numbers, why the second matrix found four new Dogs.
2. The owner's original problem was 2.5 points of food cost. List the seven possible causes from Chapter 11 and rank them by how quickly you could test each one. Which would you do first?
3. The fish-delivery collapse showed up two periods after the decision that caused it. Design a control — a report, a check, a meeting — that would have caught it in period one.
4. Was cutting six items at once ever defensible? Construct the strongest possible argument for it, then say what evidence would have been needed to make that argument sound.
5. This case is a labeled composite rather than a named restaurant. What is gained and what is lost by that choice, and what would you demand of the author before trusting the numbers?
6. Compare this case with Case Study 1. In one, a chain kept a low-margin item and worked on its structure; in the other, an independent cut its low-margin items. Both were responding to the same arithmetic. What made the responses different, and which one had the better instinct?
7. Apply the lesson to Bellwether. The ember trout is a Dog by 1.2 units a week and the only fish on the menu. Write the three-sentence decision you would put in the record, including what you are accepting and when you will revisit it.