A $3 add-on accepted on one call in four adds about $75 to a hundred-order day. Over a month of that volume it is real money, and it costs nothing but eight words at the right moment.
The same eight words at the wrong moment, or repeated to a regular who has said no every Friday since March, costs you something harder to measure. This is the tension in phone upselling, and most of the resolution is in the wording and the placement rather than in the offer itself.
The generic question earns almost nothing
"Anything else with that?" is the default line on most restaurant phones, and it functions as a closing signal rather than an offer. Callers hear it as "are we done," and they answer accordingly.
The problem is that it asks the caller to do the work. They have to remember your menu, pick something, and decide whether they want it, all while the person on the other end waits. Under that load, the easy answer is no.
A named item with a price does the work for them
"Want to add garlic knots for three dollars?" asks for a yes or a no about one specific thing. The caller does not have to recall anything, and the price removes the hesitation that comes from not knowing what they just agreed to.
Three properties make a line convert. It names one item, not a category. It includes the price. And it is short enough that the caller does not lose the thread halfway through. Anything longer than about a dozen words starts to sound like a commercial and gets refused on principle.
Relevance beats volume
The strongest upsell is the one that matches what is already on the ticket. Wings without a drink, a family-size pasta without bread, a spicy entree without something to cool it down. Each of those is a gap the caller can feel once you name it.
The weakest is the item you are trying to move this week, offered to everyone regardless of what they ordered. That version has a low take rate and a high irritation rate, which is the worst combination. If you need to move an item, put it in the greeting as a special rather than in the close as an add-on.
Where order history is available, the relevance can go further. A caller who orders the same thing every week has already told you what they want, and the useful suggestion is adjacent to that pattern rather than randomly selected. The mechanics of tying a caller to their history are covered in voice AI and loyalty programs.
One attempt, then stop
The second attempt is where phone upselling turns into telemarketing. Take rate on the follow-up is low, the caller has already declined once, and the cost lands on the relationship rather than on the ticket.
Regulars notice this faster than anyone. Someone who calls three times a week experiences your upsell prompt a hundred and fifty times a year, and a system that pushes twice every time is a system they start ordering around, usually by going online instead. That is a channel shift you pay for in delivery fees or in lost data.
Configure one attempt. If it is declined, close the order cleanly and move on.
Timing: after the order, before the total
The window is narrow and it matters. Offer too early, while the caller is still deciding, and you interrupt their own ordering. Offer after the total is read back and the transaction feels closed, so the add-on reopens something the caller thought was finished.
The spot that works is after the last item is confirmed and before the total. The order is settled, nothing has been committed to yet, and adding an item does not feel like reversing a decision. This is one of the places where a fixed sequence genuinely helps, which is the broader argument in call scripts versus natural conversation.
What to offer, ranked by how well it travels
Drinks convert best on pickup and worst on delivery, for the obvious reason that a two-liter is easy to grab on the way out and a fountain soda in a car is not.
Sides and starters are the reliable middle. They are cheap enough to be an impulse yes, they hold up in transit, and they usually carry good margin. Bread, knots, chips, extra sauce, a side salad.
Desserts are the hardest sell on the phone. The caller is imagining eating in forty minutes, at home, after a full meal, and most of them decline. Items that hold well and travel well do better than anything with ice cream in it. If dessert is a real line of business, offer it on pickup orders only.
Upgrades, meaning a larger size or a premium protein on something already ordered, are underused. They require no new decision from the caller, just a bigger version of a decision they already made, and the incremental cost to you is often lower than an added item. The broader arithmetic on ticket size is in average ticket and phone upselling.
Where the offer breaks
An upsell that names an item you are out of is worse than no upsell. The caller says yes, and then hears that it is unavailable, which is a small betrayal that lands right at the end of the call. Any add-on suggestion has to respect live availability the same way the rest of the menu does.
The other break is modifier confusion. Offering "extra sauce" when your system has four sauces and a size dimension turns a three-second yes into a twenty-second interrogation. Keep the offered item simple, or make the offered version a single default, and handle the complicated ones the way modifier handling on voice orders describes.
The number that tells you if it is working
Take rate is the obvious metric and it is not sufficient on its own. Track it alongside average ticket and alongside call abandonment in the closing sequence.
If take rate is decent and abandonment held steady, the prompt is earning. If take rate is decent and abandonment rose, you are trading completed orders for add-ons, which is usually a losing trade because a lost order costs the whole ticket. The gentler approaches in upselling without being pushy are worth reading if that is what your numbers show.
Pick one line, run it for two weeks, and write down all three numbers before you change anything. Most operators change the offer and the wording and the placement at once, then cannot tell which of them did anything.