AI Phone Agent for Quick-Service Restaurants

At 11:52 the fryer baskets are all down, there are 6 people at the counter, and the phone is ringing for someone who wants 2 number 3s, one with no mayo, and a Dr Pepper instead of a Coke.

Nobody can pick it up without turning their back on the line, and the caller is not going to wait 40 seconds for someone to finish bagging.

Quick service is built on speed at the window and a small ticket. A phone that rings 6 times at 12:10 does not feel like a missed lead. It feels like the store is slow, and the caller is 90 seconds from a competitor's drive-thru.

TL;DR

• A quick-service call is short, combo-heavy, and arrives in a burst between 11:30 and 1, so the agent has to be fast before it has to be charming.
• The combo is the product: a numbered meal with a size, a drink, and a swap, and a flattened version prices wrong even when the food is right.
• Industry estimates put roughly 30% of calls unanswered during the 5 to 8 p.m. peak, and about 85% of callers do not ring back after one goes unanswered.
• At an $11 ticket, 20 missed calls a week is about $880 a month, which is more than the Kovo Starter plan at $99.
• The order limit is the shop's own, defaulting to 10 items in total and 8 of one item; above it the call rings your handset if you asked for that, so platters stay a person's job.

A QSR phone call is 40 seconds long

Listen to 20 of them. Almost every call is one of four shapes, and 3 of the 4 are over before a full-service host would have finished saying good evening.

• The combo. Two number 3s, one large, one with a Coke, one with a Dr Pepper, no mayo on the second. That is a complete order, and it is 14 words.
• The single item. A 6-piece, a large fry, a shake. Usually a caller who is 4 minutes away.
• The status question. Is the lobby open, is the dining room closed, do you have the limited-time item, is the machine working.
• The group order. An office wants 9 sandwiches with 9 different builds, and that call needs a person or a form, not a voice agent.

The first two are worth answering automatically, because they are most of the volume and the crew cannot leave the make line to take them. The office order is the one that should reach a human, and the decline rules below are what do that.

The combo is the product, and that is where it breaks

Every vertical has a failure it does not share. A pizzeria's is the half-and-half. A QSR's is the combo, and specifically the swap inside it.

A number 3 is not a product with a price. It is a required choice group: an entree, a size, a drink, and a set of modifiers, with upcharges attached to some of those choices and not others. "Large" moves the price. "No mayo" does not. "A Dr Pepper instead of a Coke" depends on your fountain and your menu build. "Two of them, one large" is a different total for each unit.

A phone agent that treats the combo as one item with a free-text note gets the food right and the money wrong. The ticket prints as a sandwich, a fry, and a drink at full price, so the caller is quoted $9.40 and charged $12.80, or the cashier has to fix it at the window while 3 cars wait.

The fix is structural, not conversational. The combo has to exist in the menu model as a required group with priced choices, so a large upsize and a drink swap resolve to the right number before anything prints. Kovo keeps nested choice groups and reads them from the menu source, so the number the agent says is the number on the ticket. Ask for that exact test on a demo: 2 combos, one upsized, one with a swapped drink, and look at the total.

The other QSR-specific edge is the limited-time item. When the window ends, the item has to come off the phone menu the same day, or callers keep ordering a sandwich the kitchen cannot build. The menu sync re-pulls once a day and the call path only pulls when the menu is 2 days old, so an LTO that ends without a menu update stays sellable for a day. Sold out is an owner toggle in Kovo's menu board, and an owner's 86 outranks a source that still says available, so flip it there the moment the window ends rather than waiting on the pull. The one case the toggle misses is a sold-out choice, a single size of an item.

Peak throughput: what 12:10 does to a phone line

Quick service does not have a phone problem spread evenly across the day. It has two 45-minute windows, and the calls land inside them in a stack.

Industry estimates put roughly 30% of calls unanswered during the 5 to 8 p.m. peak, and lunch is no better. Roughly 85% of callers do not call back after an unanswered call, per research frequently attributed to BIA/Kelsey. In QSR terms that is not a lost reservation. It is a $9 to $14 ticket that went to the place with the app.

The crew cannot absorb the burst, and asking them to is a throughput mistake. A cashier who stops mid-bag to take a 40-second order gives back 2 minutes of queue time, because the next 3 customers all arrive at once. The phone is not competing with the counter for a person. It is competing with the counter for seconds, and the counter always wins until somebody gives up.

What a phone agent actually buys you is that the burst stops being a staffing question. Kovo picks up on the first ring, and the caller hears a menu rather than a busy signal.

The arithmetic at an $11 ticket

The numbers in quick service are small, so do the multiplication before you decide the price is too high.

Count the calls you miss in the two peak windows. Twenty a week is a conservative read of a busy store's phone log. At an $11 ticket:

• 20 missed calls a week × $11 = $220 a week.
• $220 × 4 weeks = $880 a month.

Kovo Starter is $99 with 750 connected minutes, roughly 300 calls, so the missed-call line alone pays for the subscription about 8 times, and that is before the calls you do answer, which cost the crew nothing. Calls under 10 seconds are never billed, so the wrong numbers and robocalls that make up a real share of a QSR line do not eat the month.

Now use your own ticket, because $11 is an assumption and yours may be $9 or $16. If 20 missed calls is too high, halve it and the plan still clears at 4 times. If it is too low, the case gets better.

What the agent has to get right

Five things, in the order they matter on a lunch shift.

• Speed. Turns end after 200 milliseconds of silence, and ordinary turns do not wait on a tool call, so an order moves at the pace of a speaker box rather than a call center. The end-of-turn decision is checked again before the first audio sample reaches the phone.
• One read-back, short. The order is read back before it is placed, in the order the kitchen builds it, not as a summary of the conversation.
• The ticket goes where you cook from. Toast, Square, Clover, Olo, Epson Server Direct Print, or Star CloudPRNT. A test connection prints one $0.00 ticket so the path is proven before a real order uses it.
• Payment, stated plainly. With cards off, the caller pays at the window. With cards on, a caller whose card the shop already holds is charged off-session while they are still on the line, and the ticket reaches the kitchen during the call. A first-time caller gets a Stripe Checkout link with the line held open and pays it without hanging up. No card number is spoken, recorded, or sent to a model. The gap that remains is the stranger who will not tap a link, and Loman documents capturing any card inside the call.
• Hours and the lobby. Hours come from the Google listing with a manual grid, holiday exceptions, and a last-call cutoff, which covers the 9:20 p.m. "is the lobby still open" call.

The Toast order path and the Square version describe the same mechanism from the kitchen side, if the ticket is the part you want to see before buying.

Who this is wrong for

If your phone is mostly a drive-thru lane, this is the wrong product. A lane needs a speaker box, a headset, and hardware at the window. The vendors who build that are lane vendors, and the roundup of QSR agents sorts them from the phone-only products.

If your volume is catering platters and office drops, an order-taking agent is not the buy. Those orders need lead time, a deposit, and a person. The default limits of 10 items in total and 8 of one item catch most of them, and above a limit the call only rings your handset if you turned on large-party routing; otherwise the agent says the order is too large to send through and keeps taking what fits, which is not a catering workflow.

If your menu changes weekly and nobody owns the menu data, fix that first. A phone agent is only as current as the menu it reads, and a QSR with an LTO calendar needs one person whose job includes taking the old item off both lists.

FAQ

Final Thoughts

The decision in quick service is narrow: is the phone a channel your crew can serve during the two windows that matter, or is it a queue nobody owns. If it is a queue, the fix is an agent that reads your combos correctly and prices the swaps. If it is already served, you do not have the problem this article is about.

Kovo answers on the first ring, keeps the combo structure and the upcharges intact, reads the order back, and prints it to Toast, Square, Clover, Olo, or a printer. The platter call rings your counter if you route it there. Try it for 14 days with no card, and call in your worst combo, the one with 2 swaps and a size change. Take a test call with your own menu.