The best upsell is often the item a guest would have added after more scrolling anyway — offered one click earlier. Order history is an honest map of real behaviour. Rank suggestions by contribution margin rather than clicks, then protect them with guardrails: filter allergens, remove 86'd dishes instantly, and account for each station’s load. That promotes profitable discovery rather than simply inflating the number of add-ons.
Your order history is the only honest record of how guests actually order from you. It is not a wish list; it records combinations that held up despite changes in customisation, time of day, and price. The best upsell is often already in that record: the starter that completes the main, the drink that balances the heat, or the side the kitchen can still send easily at 7:45. Often it is exactly what the guest would have added anyway — simply offered one click earlier.
The aim is not to “add something.” It is to bring one single suggestion forward from a long menu: one that improves the meal and strengthens margin. Without guardrails, “smart” suggestions become random noise that annoys guests and costs profit.
Rank by contribution margin, not clicks
A basic system always suggests €3 fries because they are popular. If margin is the goal, it needs to be smarter: score each possible suggestion by how likely this guest is to accept it in this basket — and by its contribution margin. Likelihood of acceptance times margin: that simple idea brings forward the €9 handmade starter that fits the order, instead of a €3 item the guest would have ordered anyway.
Guardrails separate discovery from chaos
A suggestion is only smart if the kitchen can deliver it. Three guardrails are essential, and all matter more than margin: if a guest has filtered for gluten-free, every suggestion containing flour disappears. If a dish is 86'd, it vanishes from recommendations that same second, not at checkout. And if the fryer has a twenty-minute backlog, the system switches to cold starters or drinks so the promised pickup time holds. Without a link to the live menu, a “smart” suggestion is just a promise the pass cannot keep.
Use patterns that are not obvious
Order history often reveals connections nobody would spot alone: guests who order extra hot are much more likely to add a particular premium water — but only when it is offered before the final step. Patterns like these are gold because they are real. Just run them through the guardrails before turning them into a suggestion.
The 7 most common mistakes
- Ranking by volume rather than contribution margin.
- Always suggesting the cheap classic.
- Letting allergen filters come after margin.
- Leaving 86'd dishes in recommendations.
- Ignoring kitchen load at each station.
- Showing too many suggestions until discovery becomes noise.
- Measuring add-on revenue only, not real margin gain.
How to find high-margin suggestions
Frequently asked questions
Why rank by margin instead of popularity?+
What are the most important guardrails?+
Can't I just show the same three add-ons everywhere?+
How do I know whether suggestions really help?+
Growth the kitchen can live with
A good upsell is not a sales trick. It is a discovery the guest enjoys and that protects your margin. Draw on real order history, rank by contribution margin, and tie suggestions firmly to the menu the kitchen can fulfil. That turns the old “Would you like fries with that?” into a precise engine that lifts the basket without burning profit. Where the suggestion appears is the next decision that shapes its effect.
