How to upsell in a restaurant without being pushy

Upsell by helping guests choose, not by pushing. Suggest one specific thing at the moment they’re deciding, describe it in a sentence, offer a choice and take no for an answer the first time. That’s what guests hear as good service. And set your expectations honestly: the best peer-reviewed study of tabletop ordering found about 1% more per check, and the popular 20% to 30% figures trace to a hypothetical example and vendor marketing. This post covers the research, the moments and phrases that work, how to train for it and how to measure it.

What does upselling mean in a restaurant?

Short answer: suggesting something the guest may want, either a better version of what they chose or something to go with it.

Two kinds get called upselling:

• Upselling: a pricier version of the same choice. A larger glass, the dry-aged steak, the premium gin.
• Suggestive selling or cross-selling: something to go with it. A side, a sauce, a drink that suits the dish, dessert.

The second kind is usually the easier one to do well, because it answers a question the guest is already asking themselves: “what goes with this?” Pushy upselling is the version where the suggestion serves the check and not the guest. Guests can tell the difference within a sentence.

Does upselling actually increase the check? What research says

Short answer: a little. Expect small, steady gains, not a jump.

Sources: Restaurant Technology News, 2019, reporting Tan and Netessine; Susskind and Curry, Cornell, 2016, p. 4; Butlaroo; AI-Waiter.com.

The study measured tabletop devices, not servers’ technique, but it is the best evidence on whether prompting guests at the table moves the check. It moved it about 1%. The vendor figures may be true for someone, but none of the pages we read gave a sample, a time period or a comparison group. We go through the evidence in full in does tabletop ordering increase check size?.

The practical lesson: upselling is worth doing well because it’s part of good service, and it adds up over a year. It is not a strategy that rescues a thin margin on its own.

Which upselling techniques work without being pushy?

Short answer: specific, well-timed, described, optional and once.

1. Name one thing. “Would you like anything else?” gets a no. “The grilled halloumi goes well with that salad” gets a decision.
2. Pair it with what they chose. A suggestion that fits the guest’s order sounds like knowledge. A random special sounds like a script.
3. Describe it in one sentence. What it is and why it’s good: “it’s charred on the grill, with honey and thyme”.
4. Offer a choice, not a yes or no. “Fries or a green salad with that?” is help with a decision the guest needs to make anyway.
5. Suggest once. If the answer is no, the subject is closed. A second try is where helpful turns pushy.
6. Read the table. A couple lingering over dinner may want dessert ideas. A business lunch watching the clock wants the check.
7. Be honest. If a guest asks whether they need the side, and the dish is a full plate, say so. Guests come back to servers they trust.
8. Never upsell around allergies. If a guest mentions an allergy, the conversation is about safety, and staff have the final say.

What should a server say? Helpful versus pushy

Short answer: the same suggestion can be either. The wording and the timing decide.

Example wording; the dishes and prices are illustrations.

The helpful column has three things in common: a specific item, a reason, and an easy way out.

When is the right moment to suggest something?

Short answer: when the guest is already deciding. Before that it’s noise; after it, it’s an interruption.

• Drinks: at the greeting, when the guest is choosing anyway.
• Starters and sides: when the main is chosen, as a choice that completes the order.
• Upgrades: when the guest hesitates between versions, not after they’ve decided.
• Dessert and coffee: when plates are cleared and the table seems unhurried.
• Never: after the check is requested, during an allergy conversation, or to a table that has said no.

How do you train staff on upselling?

Short answer: taste, pick, practise and agree when to stop.

1. Taste the menu. Servers can’t describe what they haven’t tried. Put suggested items into pre-shift tastings.
2. Pick one or two suggestions per course and write down the one-sentence description for each.
3. Practise out loud in pairs, both the suggestion and the “no problem”.
4. Agree the rules: suggest once, offer a choice, never around allergies, read the table.
5. Measure, don’t pressure. Share attach rates as a team number, not a leaderboard that rewards the pushiest server.

How can the menu do the suggesting?

Short answer: by turning suggestions into clear choices and short descriptions, so guests find them without anyone selling.

• Make sides and sizes a choice in the dish itself, with the options listed, rather than a separate section guests skip.
• Write one-line pairings under dishes: “good with the house white”.
• Describe dishes in plain words. Guests order what they understand.
• Keep the list short. Ten good suggestions beat forty that nobody reads.

This matters more when guests order from their phones, because the menu does the talking. Our guide to preparing your menu for digital ordering covers how to write choices and descriptions.

Can an AI waiter upsell? What WaiterPilot does

Short answer: WaiterPilot asks about choices and may ask once if a guest wants anything with a dish. It has no upsell rules, and we make no check-size claim.

What it does:

• Asks about required choices. If a dish needs a size, side or flavour, the guest must pick one before the order can be confirmed. That’s the “fries or salad?” question, asked every time.
• May ask once. After a guest adds a dish, WaiterPilot may ask once whether they’d like anything with it, the way a good server would.
• Answers from your menu. Guests ask questions in their own words and the answers come from your menu; when it doesn’t say, WaiterPilot tells the guest to ask a staff member.
• Leaves allergies to staff. It gives what the menu says and asks the guest to confirm with staff, who always have the final word.

In our test at BroadNet.AI’s demo café on 15 Oct 2026, the chat listed fresh juices from the menu, offered to make any of them a smoothie (an option on that menu, for 0.35 JOD), and after a juice was added, asked once whether the guest wanted anything else. That is one observed run at a demo restaurant, not a measured result.

What it doesn’t do: no upsell scripts, no pairing rules and no suggestions the owner sets up. And no claim that checks get bigger. Some vendors advertise large percentage lifts from their AI; see the table above for what those claims rest on. If your main goal is a pushier sales tool, WaiterPilot isn’t built for that. For what an AI waiter is for, read what is an AI waiter or see the AI waiter page.

How do you measure whether suggestions work?

Short answer: by attach rate on the same weekday before and after, not by the average check.

Attach rate = orders that include the suggested item ÷ orders of the dish it goes with

Casa Olivo is a fictional restaurant (an example, not a customer). The figures are assumptions to show the arithmetic.

• Casa Olivo sells 400 mains a week, and 120 of them come with a paid side: an attach rate of 30%.
• After training servers to offer “fries or a green salad?” with every main, the attach rate on the same weekdays rises to 35%: 140 sides.
• The 20 extra sides at $6 each add $120 a week.
• Against weekly sales of $32,000, that’s 0.4%.

That is roughly the scale the research suggests: real money over a year ($120 × 52 = $6,240), but small against total sales. Track attach rate by item and by weekday, and ignore the average check for this purpose; it moves with prices, party sizes and season. If your ordering tool reports top dishes, read it alongside the other numbers that matter.

What to do next

1. Pick one suggestion per course and write the one-sentence description for each.
2. Practise the wording with the team, including “no problem”.
3. Measure the attach rate for those items on the same weekdays for two weeks before and two weeks after.

To see how WaiterPilot handles choices in a real order, try a sample order. To compare AI waiter tools on what they claim and what they show, see our comparisons.