Your Best-Seller May Be Your Worst Menu Decision

Why Traditional Menu Engineering Is Not Enough

Let me start with the decision that usually creates the problem.

You run your menu-engineering report and an item appears as a Star.

It is popular. Its contribution is relatively strong. Everybody in the room can understand the conclusion quickly. “Good product. Sell more.”

That may be sensible. But I would not conclude that yet. Popularity tells me that guests are buying the item. Contribution tells me something about the economics of each sale. Neither one, on its own, tells me whether increasing sales of that item will improve the restaurant. That is the first limitation I want you to keep in mind.

A Star can still create the wrong operating result

Suppose your strongest main course sells in large volume and gives you good contribution per plate. Now suppose that during your busiest two hours it occupies the station that is already constraining the kitchen. Every additional sale uses scarce grill time. Service begins to slow, another item queues behind it and the operating team starts working around the product rather than through it. Would the item still classify as a Star?

Yes. Nothing in that description automatically makes the popularity or contribution calculation wrong.

What changes is the management question.

Instead of asking:

“Can we sell more?”

I would ask:

“What happens to the operating system if we sell more?”

That is a much better starting point.

A Star is a classification. It is not an instruction.

The matrix is useful because it simplifies the menu

I do not want to make traditional menu engineering sound weak. Its simplicity is one of its strengths. A restaurant might have twenty, thirty or fifty items. Looking at every dish individually can quickly become an unfocused discussion. Popularity and contribution give management a fast way to organise the menu into different commercial patterns.

A relatively popular, relatively high-contribution item behaves differently from a low-volume, low-contribution item. That is useful.

The problem begins when management moves directly from classification to prescription.

The four-box matrix is very good at saying:

“Look here.”

It is much weaker at saying:

“Do this.”

That is why this guide preserves the classical screen instead of replacing it. SCREEN is the first stage of the TRAIL method, but it is deliberately treated as orientation rather than instruction.

Take the Plowhorse

A Plowhorse is popular but generates relatively lower contribution.

The obvious management reaction is often:

“Raise the price.” Maybe. But why is the contribution low? Perhaps the item really is underpriced.

Perhaps the recipe cost has moved and nobody updated the selling price. Perhaps the portion has become larger than the intended specification. Perhaps discounts are pulling the net selling value down. Perhaps the product is deliberately positioned as an accessible entry item.

Perhaps the guest who buys it also buys a drink or side with very attractive economics. Or perhaps the item is already doing an important commercial job and the better decision is to improve recipe architecture rather than put demand at risk. Those situations all produce the same broad classification. They do not justify the same action.

Now take the Puzzle

A Puzzle produces relatively good contribution but lower popularity.

A common response is:

“Promote it.” Again, maybe. But what if the low demand exists because the price-value relationship is wrong? The item may have attractive contribution per plate precisely because the selling price is high. If that price is suppressing demand, spending more money to promote the product may simply push more marketing behind the wrong proposition.

The problem could also sit somewhere else. Perhaps the item is frequently unavailable. Perhaps the description is unclear. Perhaps the preparation is difficult and the service team does not recommend it enthusiastically because they know what will happen in the kitchen.

Perhaps the item is behaving exactly as management intended because it is a premium niche choice rather than a volume product. One classification. Several possible explanations. Several very different decisions.

And then there is the Dog

The Dog is where automatic menu engineering can become particularly dangerous. Low popularity. Lower contribution. Remove it.

That conclusion sounds logical until you ask what disappears with the item. Suppose it is the only credible vegan main course. Suppose it provides an important dietary option. Suppose it is a local signature item.

Suppose it gives the menu an important lower or higher price point. Suppose it is required for a package or a specific guest occasion.

Its standalone sales may be weak while its line-up role remains important.

That does not mean the item deserves permanent protection. A weak vegan curry can still be replaced with a better vegan main course. A local signature dish can still be redesigned. A product can keep the same role while the SKU changes.

But the decision has changed from:

“Remove the Dog.”

to:

“If I remove this item, what do I need to replace so the menu still performs its intended role?”

That is a much stronger management conversation. Later in the guide we will compare two Dogs from the worked example: the Vegan Curry, which protects a required menu role, and the Imported Chocolate Tart, which does not protect anything equally unique. The matrix classifies both as Dogs, but the wider evidence points management in different directions.

Same quadrant, different role, different management decision.

A monthly result can also be temporary

There is another weakness in using the quadrant as the answer. The quadrant itself can move. Suppose an item has been a Puzzle for several months and then suddenly becomes a Star. What changed?

A promotion may have been running. A food festival may have shifted demand. A competing item may have been unavailable for part of the period. The menu position may have changed.

The recipe may have changed. The price may have changed. The hotel may simply have had a different guest mix. The current classification can therefore be mathematically correct while still being a poor basis for a permanent menu decision.

The opposite can also happen. A seasonal item can look weak when averaged over the full year but perform exactly as intended during the season when the item matters. I do not want you to stop reviewing the menu monthly. A monthly movement is useful because it tells you that something changed.

I want you to separate the review frequency from the decision horizon.

Review regularly. Make structural decisions from representative evidence. The Seafood Plate worked example later in the guide will show this clearly: the item moves from Puzzle to Star during a specifically identified promotional period and then returns to Puzzle. The Star classification is real for that month, but it does not establish that the item permanently became a Star.

A promoted month can be real and still be non-representative of normal demand.

The item may be profitable and still be a poor use of capacity

There is also a cost question that the classic matrix does not fully resolve. Traditional contribution is normally built from the item’s selling economics less product cost. That is useful and we will calculate it properly in the next chapters. But suppose two dishes both generate strong contribution.

One occupies the constrained grill for ten minutes. The other occupies it for fourteen. If the grill has plenty of available capacity, the difference may not matter very much. If the grill is already the bottleneck at peak dinner, the difference becomes commercially important.

The second item may give more contribution per plate and less contribution from each minute of the scarce resource. That does not automatically make it a bad product. It simply gives management another piece of evidence that the four-box model does not contain.

This is the reason TRAIL separates Retained Economics from Activity & Capacity rather than trying to force every operational issue into one contribution number. The framework limits bottleneck-minute economics to situations where a genuine constrained resource has actually been identified.

The item may also affect other items

The same issue applies to what the guest buys around the item. A Club Sandwich may frequently appear on the same check as a Local Beer. A premium Seafood Plate may have a strong association with a glass of wine. If you reprice, remove or promote the focal item, part of the commercial effect may appear somewhere else on the check.

There is another side to this. If Seafood becomes less attractive, some guests may choose Grilled Chicken instead. If you remove one dessert, some demand may shift to another dessert rather than disappear. The item is therefore not always the complete unit of economic analysis.

Sometimes I need to look at the basket, the category or the system effect.

That does not mean every association is causal. The framework is explicit about that boundary. Items appearing together does not prove that one caused the other to be purchased, and associated contribution must not be added again as if it were new restaurant revenue. The relationship is evidence to investigate and, where worthwhile, test.

So what is the matrix actually telling me?

By now, I hope the distinction is becoming clear. When an item appears as a Star, Plowhorse, Puzzle or Dog, I do not want the discussion to end. I want it to begin.

A Star asks:

Why is this working, and what should I protect before I try to increase it?

A Plowhorse asks:

Why do guests want this item while its contribution remains relatively weak?

A Puzzle asks:

Why does an economically attractive item attract relatively less demand?

A Dog asks:

Is this genuinely a weak item, or is it performing another required job on the menu?

Those questions preserve the value of traditional menu engineering without allowing the label to make the decision for us.

What I recommend instead

The solution is not to replace one simple model with a more complicated automatic model. I do not want TRAIL to become another spreadsheet that tells you what to do. The companion workbook deliberately calculates and flags evidence but leaves the management action open.

The sequence is:

SCREEN the menu using the classical model.

Then follow the priority item through:

Trend & Target

Retained Economics

Activity & Capacity

Interactions

Line-up Role

Then TEST the proposed intervention.

The test matters because even a very good diagnosis is still a hypothesis until you change something and observe what happens.

SCREEN tells you where to look. TRAIL builds the decision. TEST verifies it.

What I want you to stop doing

Before we move into the calculations, I want you to make one behavioural change. Stop translating the quadrant directly into an instruction.

Do not say:

“Star — promote.” “Plowhorse — increase the price.” “Puzzle — promote.” “Dog — remove.”

Instead say:

“This classification tells me where I need to investigate.”

That single change protects you from many of the weak decisions we will address throughout the rest of the guide.

Before you move on

Take one menu item that management has recently discussed changing. Write down its current classification if you already have one. Then, without making a decision yet, ask five simple questions: Is the current result representative?

What does the item really retain? What operational load does it create? What does it interact with? What role would disappear if I removed it?

If you cannot answer those questions yet, that is fine.

You have already learned something useful:

the quadrant is not enough evidence to act.

In the next chapter, I will go back to the classical menu-engineering model and show you exactly what it still does very well, how we calculate it, and where I want you to stop before the label becomes the decision.

CHAPTER 2