AI recommendations
How MenuWright's AI generates prioritized, dollar-quantified actions for each menu item.
How it works
For each menu item, MenuWright's AI (powered by Claude) analyzes:
- Matrix classification (Star / Crowd Pleaser / Hidden Gem / Leftover)
- Margin and popularity trends over the selected period
- Category context (how the item compares to its peers)
- Bundle and cannibalization signals
The output is a prioritized action with a projected monthly dollar impact.
Recommendation types
| Action | When | Example |
|---|---|---|
| Promote | Hidden Gem — profitable but under-ordered | "Move to the top of the entrée section; projected +$420/mo" |
| Reprice | Crowd Pleaser — popular but under-margining | "Increase price $1.50; projected +$310/mo at current volume" |
| Bundle | Complementary items with co-ordering signal | "Bundle with the house salad; projected +$180/mo" |
| Reduce cost | High-volume item with margin pressure | "Switch to a lower-cost protein; projected +$250/mo" |
| Cut | Leftover — low margin, low volume | "Remove; frees menu space for a higher-performing item" |
| Protect | Star — already optimal | "No change needed; monitor for margin erosion" |
Confidence and caveats
Each recommendation carries a confidence level (high / medium / low) based on data volume and trend stability. Low-confidence recommendations appear with a warning badge.
AI recommendations are decision support, not directives. They are based on historical sales data and do not account for factors like chef expertise, supplier relationships, customer sentiment, or seasonal events you may be planning. Always apply your own judgment before changing your menu.
Applying recommendations
Recommendations are suggestions — MenuWright does not modify your POS or menu. Track which recommendations you have acted on using the Status toggle (pending / applied / dismissed) on each item card. Applied recommendations feed into the next analysis cycle so the AI can measure actual impact.