Three weeks after a pizza shop turned on a single side prompt, garlic knots went from roughly one in eight phone orders to closer to one in four. Average ticket was up about two dollars. The owner was pleased, right up until the Friday the fryer station started running four minutes behind on every ticket that had knots on it.
That is the whole subject in one story. Upsells do not just make tickets bigger. They change what you sell, and what you sell is what your kitchen has to produce.
The shift shows up fast and lands in one place
An upsell prompt is not a gentle nudge that accumulates. It runs on every qualifying call starting the day you turn it on, which means the effect is close to immediate and it concentrates almost entirely in the category named by the prompt.
If the agent offers a side, side attachment moves. Drinks and desserts usually do not, unless they are also prompted. That concentration is the single most useful property of the whole thing, because it gives you a built-in control group. When one category jumps and its neighbors sit still, the prompt caused it. When everything drifts together, you are looking at weather or a holiday or a promo running somewhere else.
Pull your item mix for the two weeks before and the two weeks after. Not revenue, not ticket average, item counts as a share of orders. That is the report that shows you what actually happened.
The good version and the expensive version
There are two ways average ticket goes up, and they are not the same thing at all.
The good version is addition. A caller ordering two entrees now also orders a two-liter and an order of knots. Nothing about their original intent changed, the ticket got larger, and the incremental items are cheap to make and high margin. This is what a well-built prompt does and it is why upselling on the phone is worth doing.
The expensive version is substitution. The agent frames a premium item in a way that pulls the caller off what they called to order, and the ticket rises while the customer's satisfaction with the order falls. You do not see this in the mix report for a month, and when you do see it, it looks like a category winning. The tell is in the other direction: the category it won from should be visibly shrinking.
If your prompted category is up and an adjacent, more ordinary category is down by a similar count, you are substituting, not adding. That is a prompt to rewrite, and upselling without being pushy covers the wording that avoids it.
Kitchen load is the constraint nobody budgets for
A prompt that attaches an appetizer to a quarter of your phone orders has done something to your production plan, and nobody told the kitchen.
The math is plain. If you run 120 phone orders on a Friday and the prompt lifts fried-appetizer attachment by fifteen points, that is roughly eighteen extra fry orders concentrated in the same three hours as everything else. One station absorbs all of it. Ticket times on anything touching that station will stretch, and the stretch will show up on dine-in tickets too, because the fryer does not know which channel a ticket came from.
Four things to check in the two weeks after you enable any prompt:
- Ticket time on the station that produces the prompted item, compared against the same station in the prior two weeks at the same daypart.
- Whether the prompted item's waste or hold time changed, since a batch item held longer to keep up with demand is a quality problem before it is a cost problem.
- Whether pickup time quotes are still accurate, because a quote built on your old production reality will now be short. Quoting accurate pickup times is the fix if it drifts.
- Whether any prompted item has gone out of stock more often, which is a signal to tighten your 86 sync before the agent starts selling something you cannot make.
None of this is an argument against upselling. It is an argument for turning prompts on one at a time and watching the kitchen for a fortnight, rather than enabling six of them the week you go live.
Prompt on cheap, fast, complementary items
The categories that behave well share three properties. They are inexpensive relative to the entree, so the caller does not have to make a real decision. They are fast to produce or already produced, so attaching them does not lengthen the ticket. And they genuinely go with the order, so the offer reads as helpful instead of transactional.
Drinks are the cleanest example. A two-liter or a canned soda adds to the ticket, costs the kitchen nothing, and nobody has ever been annoyed at being asked. Desserts are close behind, particularly anything held ready. Sides work well but carry the production load described above, so they need watching.
What performs poorly is anything that asks the caller to reconsider their order. Premium upgrades, second entrees, and combo restructuring all read as pressure over the phone in a way they do not on a screen, because the caller cannot skip past them. The conversion is low and the annoyance is real. The broader effect on ticket size is laid out in average ticket and phone upselling.
Reading the report without fooling yourself
Two comparisons do most of the work.
First, prompted category against unprompted category, over the same weeks. This separates your prompt from the world. If desserts jumped and everything else held, the prompt did it.
Second, item counts against order counts. A category can grow in revenue purely because you took more orders that month. Attachment rate, meaning the share of orders containing at least one item from the category, is immune to that and it is the number to track. It also survives menu price changes, which raw category revenue does not.
Watch both for four weeks, then decide. If attachment rose, the kitchen absorbed it, and the substituted-from categories held steady, the prompt is earning its place and you can consider a second one. If attachment rose and a neighboring category fell by roughly the same amount, you moved sales sideways and added work. If attachment rose and station ticket times went with it, the prompt is profitable on paper and expensive in the dining room.
The one thing not to do is turn on three prompts at once in month one. You will get a bigger ticket average and no idea which prompt bought it, which station is straining, or what to switch off when Friday goes sideways.