AI-assisted menu optimisation: choose your next change

The menu affects purchasing, kitchen work and guest choices. Menu optimisation works best when sales and costing are considered alongside experience from service. Start with a decision you need to make, such as which dish to develop for the next menu period.

Build a comparable dataset

Choose a period representative of the service you want to improve. Compare starters with starters and mains with mains. Note whether an item was sold out, newly introduced or only available on certain days. Low sales can reflect limited opportunity to order it.

Collect portions sold, sales excluding VAT and current ingredient cost per portion. Recipe costing should account for preparation yield and normal portion size. The POS records sales; it does not automatically know a new purchase price or what the kitchen discards.

Consider contribution per portion and the total

In a basic recipe calculation, contribution per portion is the selling price excluding VAT minus ingredient cost. If you also deduct packaging or other variable costs, apply the same method to every item compared. The result is not net profit.

A hypothetical example shows why volume matters:

Dish B contributes more per portion, but A contributes more overall during the period. Before steering guests from A to B, consider demand and kitchen capacity too.

Use the matrix to ask questions

Classic menu engineering combines sales volume and contribution, an approach included in Cornell’s restaurant revenue management programme. Define and document the high/low thresholds for the menu category you are analysing.

• Stars: high popularity and high contribution per portion. Protect quality and availability.
• Plow horses: high popularity and lower contribution. Review recipe, yield and price while preserving what guests value.
• Puzzles: lower popularity and high contribution. Check whether the name, description or staff knowledge needs improvement.
• Dogs: lower popularity and lower contribution. Consider the item’s role before replacing it.

An item may matter as a vegetarian option or part of the restaurant’s identity. A matrix label starts a discussion; it does not automatically justify removal.

Use AI to support investigation

For example, ask the assistant to summarise which mains lost sales share across comparable weeks. Then ask which information is missing to assess why. A useful answer separates observations from possible explanations.

Historical sales do not establish exactly how guests will respond to a new price. Weather, guest volume, other menu changes and availability may have influenced earlier results. Check calculations and the original report before acting.

Run a test you can evaluate

Choose one change, such as a clearer description for a puzzle. Define a comparable test period and monitor both the dish’s contribution and the whole category’s result. Growth in one item may come from guests switching away from another.

Discuss waiting times, waste and guest feedback with kitchen and service staff. Record what you retain and what you reverse.

Vendion Analytics provides sales information, while the AI assistants can support questions and summaries. The combination is useful when costs are current and the next decision is clear.