Every operator I know has the same line in their training binder: offer the add-on. Ask if they want to make it a meal. Mention the catering trays. And every operator I know also knows how often that actually happens on a busy Friday at 12:15. Not often.
My brother Rohit and I run our own stores in San Diego, and the upsell was one of the things we least expected to change when we put an AI on the phone. We built Maya to stop calls ringing out. What we noticed afterwards was that the orders she took were bigger. This post is about why that happens, what a natural upsell sounds like compared with a scripted one, and how to work out what it is worth to your stores, with your own numbers.
Why the human upsell is inconsistent
None of this is a complaint about staff. The people on our counters are good at their jobs. The problem is that the upsell is the first thing to go when a person is under pressure, and the phone is where the pressure is worst.
Think about who answers the phone during a rush. It is whoever is closest, usually someone who was in the middle of making a sandwich, with a line at the counter and a ticket printer going. Their goal on that call is to get the order down correctly and get off the phone. Asking "would you like to add avocado to that" is a second question, and a second question is thirty more seconds they do not have. So it gets skipped. Not on every call, but on enough of them that the upsell rate on the phone ends up depending on who picked up and how busy the store was at that moment.
Then there are the calls nobody picks up. Roughly 30% of our inbound calls were going unanswered before we fixed the phone, almost all of them in the lunch and dinner rush. An order you never took has no add-on on it either. The upsell problem and the missed-call problem are the same problem viewed from two sides: the phone rings hardest exactly when your team has the least attention to give it.
Natural, not scripted
The obvious answer is to automate the question. The obvious mistake is to automate it badly. A phone system that reads out the same promotion no matter what you ordered is not an upsell. It is an interruption, and customers tune it out or get annoyed by it.
When we were building Maya, the rule we set was that the offer has to make sense for the order in front of her, and it has to come at the point in the conversation where a person would naturally say yes. A few examples of what that means in practice:
- The modifier that fits. Someone orders a turkey sandwich; Maya asks whether they want avocado on it, the same way a good counter person would. Someone orders a steak sandwich; she offers double meat. She does not offer both to everyone.
- The side at the end. Chips, a drink or cookies get offered once the main order is confirmed, not halfway through it, so the caller is not juggling two decisions at once.
- The catering bundle. When a caller is ordering for a group, the useful suggestion is not a cookie. It is the tray, the drinks and the dessert platter that round out the order, plus a mention that the store does catering if they are planning another office lunch.
- Taking no for an answer. If the caller declines, she moves on. One offer, asked well, beats three offers that make the customer feel worked.
Each store's Maya is set up on that store's own menu, pricing and upsell sequence, so the offer is always something the store actually sells at the price it actually charges. And because she is not making sandwiches at the same time, she asks on the 12:15 call exactly the way she asks on the 3:30 call. That consistency is most of the effect. She is bilingual in English and Spanish, so the Spanish-speaking caller hears the same offer, not a shortened version because the person who picked up was not comfortable switching languages.
What it is worth: the per-order arithmetic
Here is where I have to be straight with you. In the first post we wrote about Maya, I said average order value went up in our stores without us doing anything else differently. That is true. What I do not have is a clean, isolated percentage for the upsell on its own, separate from the extra calls we started answering and everything else that changed at the same time. So I am not going to give you a lift figure. I would rather show you the arithmetic and let you run it with your own menu.
The formula is simple:
Added revenue per week = phone orders per week × share of orders that take an add-on × average add-on price
And a worked example. Every input below is a made-up illustration, not a measured result — swap in your own:
- One store takes 200 phone orders a week.
- Today, in a rush, staff remember the add-on on some of those calls. Say a consistent offer gets one more order in five to take something it would not have otherwise: 40 extra add-ons.
- Say the average add-on is $2 — avocado, double meat, a drink or chips.
- 40 × $2 = $80 a week, about $4,160 a year, for one store, from orders you were already taking.
That is the small end. The larger number is catering. Our average catering order is $500+, and a group order is where a bundle suggestion moves the ticket most, because the add-on is a tray or a drinks package rather than a single side. If you want to illustrate it, take one catering order a week and imagine a $40 drinks-and-dessert add-on landing on it. That is another $2,080 a year per store. Again, an illustration: use your own catering menu and your own prices.
Multiply either line by your store count. The upsell is not a big number on a single order. It is a small number that, asked consistently, lands on a large share of orders, every week, at every store.
For context on the whole model rather than the upsell alone: across our own stores, sales are up about 40% on this model. That figure is everything together — answering every call, catering recovery, reviews — and I would not attribute it to upselling by itself. It is also our stores, not a promise for yours.
How to check it is working
The thing I like most about having the phone on Maya is that I do not have to take the upsell on faith. Every call has a full transcript and recording, so you can read exactly what she offered, whether the customer said yes, and what the order came to. If an offer is landing badly, you see it in the transcripts and change the sequence.
Two practical notes. First, Maya does not push orders into your POS. She takes the order down, including the add-ons, and texts it straight to the store so your team only has to make it. Second, she is not a replacement for the person who is great on the phone. When a caller wants a human, or a large catering order needs a personal touch, she warm-transfers to your manager and introduces the situation before bridging the call.
The short version
The upsell on the phone fails for the same reason calls go unanswered: the phone is busiest when your team is busiest. An AI on the phone fixes the coverage first, and the consistency of the offer comes with it. It will not double your ticket, and anyone who tells you a specific lift figure for your stores without seeing your numbers is guessing. But a modest add-on, offered well on every order, is real money you are currently leaving on the counter.
If you want to see what this costs per store, the plans are on the pricing page. And if you would rather hear it than read about it, book a call and we will walk you through how it runs in our stores.