menu-engineering-quadrant

---
name: menu-engineering-quadrant
description: Sorts every dish into stars, plowhorses, puzzles or dogs by plotting popularity against contribution margin, using the method developed by Michael Kasavana and Donald Smith, and gives the standard action for each quadrant rather than leaving the chart to speak for itself. Needs real sales data and a real recipe cost per dish. Use at a fixed review interval, monthly is typical, never on a newly launched dish.
---

# Where every dish sits: sells well or not, makes money or not

Give this a period's sales count per dish and its contribution margin (menu price ex VAT, minus food cost). Get back each dish's quadrant and the standard action.

## What it does

1. **Use a real sales period, not a guess.** Popularity is "units sold divided by total covers, expressed as a percentage", over a fixed window, "four to eight weeks" typically. Two weeks is no popularity figure worth plotting.
2. **Use contribution margin, not food cost percentage, as the profit axis**: "menu price (ex GST) minus food cost per serve." A high food-cost dish can still carry a healthy margin if priced right; percentage alone hides that.
3. **Plot each dish against the menu's own average on both axes**: a Star sells above average popularity and margin; a Plowhorse sells above average popularity but below-average margin; a Puzzle sells below average popularity but above-average margin; a Dog falls below average on both.
4. **Give the standard action per quadrant.** Stars: "Keep them consistent, promote them as much as possible." Plowhorses: reduce cost to prepare, "swapping expensive ingredients, pairing... or decreasing portion size." Puzzles: find out why it isn't selling, "poorly promoted or priced too high", before cutting it. Dogs: "re-inventing... or making them less visible."
5. **Never move a dish on one period alone.** It's a repeat review, not a one-off verdict; a bad month for a Star is a reason to check for a cause, not reprice on the spot.

## Then it checks

1. Every dish plotted has a sales count and margin from the same stated period.
2. The menu average for both axes is calculated from the actual dish set, not assumed.
3. Every placement matches the averages shown.
4. The recommended action matches the fixed action for that quadrant.
5. Any dish newer than the review window is excluded and listed separately, not forced into a quadrant on partial data.

Any check fails: redo once. Failed twice: say what is wrong and stop.

## Rules

- Public information only. Never invent a fact, number or quote, or a sales figure or margin; both come from real records.
- Anything sent in someone's name says whose name it is.
- Never plot a dish with less than the minimum review period of sales history.
- Never recommend cutting a Dog on its first appearance without checking it isn't seasonal or recently launched.
- Working document for the owner to check against their till data and recipe costs, for a manager to decide what changes. Not a substitute for that decision.

## Built from
- Loaded, "The Menu Engineering Matrix Explained", https://www.loadedhub.com/resources/menu-engineering-matrix-restaurant, copyright 2025, read 4 November 2026: Kasavana/Smith attribution, quadrant definitions and formulas, steps 1-3.
- Craftable, "What stars on your menu push popularity + profitability", https://craftable.com/blog/what-stars-on-your-menu-push-popularity-profitability-with-menu-engineering, published 13 April 2020, read 4 November 2026: the action per quadrant, step 4.

# menu-engineering-quadrant

## You are given
A review period (start/end date, at least four weeks). For every dish during that period: units sold, total covers, VAT-exclusive menu price, and food cost per serve (ideally from dish-cost-builder). The launch date of any dish, to exclude ones newer than the review period.

## Produce
Write into a `./menu-engineering-output/` folder:

1. `dish-metrics.csv`: `dish_name`, `launch_date`, `eligible_for_review` (`yes` only if `launch_date` precedes the review start), `units_sold`, `total_covers`, `popularity_percent` (`units_sold / total_covers * 100`), `menu_price_ex_vat`, `food_cost_per_serve`, `contribution_margin` (`menu_price_ex_vat - food_cost_per_serve`).
2. `menu-averages.csv`: `review_period_start`, `review_period_end`, `average_popularity_percent`, `average_contribution_margin`, `dishes_included`. Averages use only `eligible_for_review: yes` rows.
3. `quadrant-placement.csv`: `dish_name`, `popularity_percent`, `contribution_margin`, `quadrant` (`star`/`plowhorse`/`puzzle`/`dog`, from comparing the dish to `menu-averages.csv`), `recommended_action` (fixed text: stars `keep consistent, promote`; plowhorses `reduce cost to prepare without cutting popularity`; puzzles `investigate low sales before cutting, check promotion and price`; dogs `re-invent or make less visible on the menu`).
4. `excluded-dishes.csv`: `dish_name`, `launch_date`, `reason_excluded` (`launched after review period start, insufficient sales history`).
5. `gaps.md` - numbered list of: every excluded dish; every dish with no food cost per serve; any dish newly placed in `dog` this period; the count reviewed versus excluded.

## Rules
- Codex calculates popularity and margin, compares to the menu's own average, and assigns the fixed quadrant and action. It never invents a sales figure, cost, or placement not derived from the data.
- A dish launched inside the review period is excluded entirely, never force-fitted on partial data.
- `recommended_action` is always the fixed wording for that quadrant; Codex never writes a custom recommendation.
- A dish newly appearing in `dog` this period is named specifically in `gaps.md`, since a human check is required before action.
- British English, GBP, DD Month YYYY dates. No em dashes.
- Every file ends with: this is a working document prepared for the owner to check against their own till sales data and recipe costs, and for a manager to decide what actually changes on the menu. It is not a substitute for that decision.

## Return
The absolute path and row count of each file. The review period and averages used. Every dish's quadrant and action. Every dish excluded, and why. Any dish newly placed in `dog`, named specifically. The `gaps.md` item count.