Restaurant Upselling: How to Train It Without the Sales Pitch

By Terry Psaltakis, Founder, ShiftTrained

Why “would you like dessert?” does not work

It is a yes or no question asked at the moment a guest feels full, and it carries no information. The guest has nothing to picture, so the cheapest answer is no. Most upselling training tries to fix this by working on the ask: be more confident, suggest rather than offer, use assumptive language. That is coaching the wrong half of the sentence.

Compare it with a server who says the pastry chef makes the tart in-house every morning and it comes warm with brown butter ice cream. Nobody had to be talked into anything. The guest was given something to want. The difference between those two servers is not personality or nerve, it is that one of them knows the dessert menu and one of them does not.

Upselling is a knowledge problem wearing a sales problem's clothes

A server who is unsure what is in a dish avoids recommending it. A server who is unsure what something costs steers to the item they are certain about. This is rational behaviour: nobody wants to describe a dish wrongly to a guest, and nobody wants to be corrected at the table. Uncertainty does not produce a bad recommendation, it produces no recommendation at all, and that is invisible on every report you have.

Which is why upselling scripts tend not to survive contact with a floor. A script assumes the knowledge is already there and only the phrasing is missing. On most floors it is the other way round.

The four add-ons that carry most of it

Entrees mostly sell themselves, because the guest arrived intending to buy dinner. The revenue that actually depends on your staff sits around the edges of the meal. These are median menu prices, sampled across 110 restaurants and 14,242 menu items:

Dollar-denominated listings only. Every figure carries its sample size on the full category table.

How to train it in a week

1. Pick four items, not forty. One dessert, one cocktail, one wine, one starter. A floor that can sell four things well beats a floor that has skimmed everything.
2. Write one sentence for each. What it is, what makes it worth ordering, and one detail nobody would guess. That sentence is the whole upsell.
3. Quiz it, do not brief it. Pre-shift briefings evaporate by table three. Being asked a question and getting it wrong is what makes it stick.
4. Include the price. Price recall is the most commonly skipped part of menu training and the one that most directly changes whether a server recommends confidently.
5. Rotate weekly. Four new items a week is a trained floor inside two months, without a single extra meeting.

How to tell whether it is working

Average check is the obvious measure and the slowest to move, because it blends everything. Watch attachment rate on the four items you trained instead: what share of covers included one. That number responds within a week and tells you which of the four the floor actually absorbed.

Restaurants that trained their floor on the menu saw check totals up 11% and wine sales up 34%, loaded from real POS data across our customers, with no menu changes, no price changes and nobody hired. The full field report shows how it was measured. If you want to size it against your own floor, the upselling calculator runs the arithmetic on your staffing.

About the Author

Terry Psaltakis is a 30-year restaurant operator who has opened more than 20 concepts across multiple markets, in every role from dishwasher to Owner.  He founded ShiftTrained in Chicago to solve a problem he lived for three decades: pre-shift meetings don't actually train staff.  Terry writes about the operational side of restaurant training, AI in hospitality, and what works on the floor.

LinkedIn · terry@shifttrained.com

Last reviewed June 2026