AI Phone Agent for Ghost Kitchens

A ghost kitchen is a kitchen with no front of house, which means the phone has nowhere to ring. There is no host stand, no counter, no cashier with a free hand. There is an expediter, a wall of tablets, and a printer.

So the call goes to whoever has a phone in their pocket, usually the operator, usually mid-service, usually with a courier at the door asking for an order number.

The result is predictable. The line is answered when it can be and dropped when it cannot, and a dropped call at a delivery-only brand is not a customer who walks in later. It is a customer who opens an app.

TL;DR

• A ghost kitchen has no front of house, so the phone lands on whoever is expediting, and it is the first thing dropped.
• The failure that belongs to virtual brands is the wrong brand answering: three menus on one number will eventually quote the wrong one.
• Kovo runs one store, one line, one menu per subscription, so three brands is three numbers, not one clever agent.
• Kovo does not dispatch to DoorDash, Uber Eats, or Grubhub. Delivery is the shop's own driver; a marketplace order belongs in the app.
• Three brands on Starter cost $297 a month, and one recovered $32 phone ticket a week covers a line.

A ghost kitchen's phone has no room to answer it

Most restaurants at least have a person near a phone. A virtual brand has a production line.

The staff are packing, labeling, and sealing. The tablets are the revenue. The phone is a channel with no owner, so the call gets answered by the same person who is bagging, which means it gets answered badly or not at all.

That is not a training problem. It is a staffing geometry problem, and it is why a phone agent fits this room better than most.

The failure that belongs to ghost kitchens: the wrong brand answers

Three brands, one kitchen, one number. A caller dials the number printed on the bag, asks for a large pepperoni, and gets the wing menu read back, because that number belongs to the wing brand.

Or worse: the operator gives all three brands to one agent, the menus overlap in the catalog, and the ticket prints a pizza under a burger brand's name at a price from neither. The kitchen builds what the paper says. The guest opens the bag and finds the wrong food, and there is no server to apologize.

That is the failure no other vertical has. A pizzeria's wrong order costs a remake. A virtual brand's wrong order costs the brand, because the only thing a virtual brand owns is its name.

One number per brand, on purpose

Kovo's structure forces the right answer here. One store, one line, one menu, one subscription. A second brand is a second number, a second menu, and a second $99.

Operators hear that as a tax. It is closer to insurance. The wing brand cannot quote a pizza price because it does not have a pizza menu, and a caller who dials the wrong brand hears the correct one tell them so.

The menu somebody has to keep

A ghost kitchen usually has no POS catalog, because it was never a counter. Kovo takes one menu source per shop: a POS catalog, or a menu page, or a pasted link. A DoorDash or Uber Eats listing page works as that source, which is the practical path for a delivery-only brand.

Marketplace pages carry marketplace prices, so Kovo's menu markup percent can lift an auto-imported menu's prices to your own. The three-way merge keeps your edits through refreshes, and an item you delete is not resurrected the next morning. The menu sync mechanics explain the rules.

The courier problem

This is the part of the page most ghost-kitchen articles skip. Kovo does not dispatch to DoorDash, Uber Eats, or Grubhub.

So a phone order at a delivery-only brand is a pickup order. There is no "hand it to the marketplace" step, no delivery address on the call, and the caller who expects a courier at the door at 7:30 will not get one.

That leaves three honest ways to run it:

• Pickup only. Fine at a commissary with a door and a shelf, thin for a brand whose guests know it only as an app tile. It works because the phone still fills the ticket the tablet would have taken.
• A driver you already run. Works if you have one, and the phone order is still a pickup ticket: the driving is your operation, and the agent takes no address and quotes no delivery time.
• Set the expectation on the call. Say pickup, name a window, and let the caller decide.

What you should not do is buy a phone agent expecting it to feed the same courier network the tablets feed. It will answer, take the order, and print the ticket. The last mile is yours, and the call is where you say so.

The arithmetic, per brand

Ghost kitchens have to do this per brand, because nothing pools.

• One brand: 8 missed calls a week × $32 average ticket = $256 a week.
• $256 × 4 = $1,024 a month, on one brand.
• Three brands on Starter = $297 a month.

That looks like an easy yes, and it usually is, with one caveat: the $1,024 assumes those callers would have ordered at full price and would not have used the app instead. Most of them would have used the app, and paid the commission. Third-party delivery commissions run 15–30% per order, toward 30–40% all-in with fees and promotions, according to industry estimates collected by the Independent Restaurant Coalition. A phone order keeps that commission, which is the actual argument for answering.

Kovo Starter is $99 per store. Unlimited is $299. If one brand takes most of the calls, put that one on Unlimited and leave the others on Starter.

The order limit and the party platter

The limit belongs to the shop: 10 items in total and 8 of one item for a brand that never opened the card, with an owner free to change either, add an order-value limit, or set one to zero to remove it.

At a ghost kitchen, the default catches the catering inquiry, the big family order, and the office lunch. Those are also the orders with the best margin and no commission, so the limit matters more here than at a restaurant with a walk-in trade to fall back on.

Above a limit, the call rings the handset if you turned on large-party routing and set a handset number; otherwise the agent says the order is too large to send through, stays on the call, and keeps taking what fits. Practically: raise the item limit to the catering you actually take, or keep one phone number, the operator's, for anything over it and say so in the greeting.

Payment when there is no counter

A delivery-only brand cannot say "pay when you collect" and mean it the way a pizzeria does. There is no counter, and often no guest, just a bag on a shelf.

So cards on is the normal configuration, and the useful path is the repeat caller: a card the shop already holds is charged while the caller is on the line, so the kitchen builds a ticket that is already paid. A first-time caller gets a Stripe link with the line held open, and the ticket waits for checkout.session.completed; if they never pay, nothing prints, which is the right default for a bag on a shelf.

No card number reaches the call, the recording, the transcript, or the model: the known-card charge uses the vaulted card and the link is typed on Stripe's page. The Loman comparison is where the difference between a known card and any card is spelled out.

The ticket goes to Toast, Square, Clover, Olo, or a printer. If the kitchen runs off one tablet, the Square route is the shortest path from a ring to a printed ticket.

What to test, brand by brand

• Call the pizza brand's number and ask for wings. Confirm the answer is "wrong brand," not a wing ticket.
• Order the item with the most modifiers in your catalog. Check the ticket structure, not the transcript.
• Push one order past your item limit. Confirm what your setting does: prints, declines, or rings your handset.
• A card order. Confirm nothing prints until the link clears, and that a second call from the same number is charged without one.
• A location question. Confirm the agent says pickup, since there is no dining room to describe.

That last one is the test most ghost kitchens skip and the one their callers notice. The missed-call math covers the volume side of the same decision.

Who a ghost kitchen phone agent is wrong for

If nearly all of your revenue comes from the tablets and your phone rings twice a week, $99 a brand is $99 too much. Fix the tablet flow first.

If your calls are courier complaints, an agent will not help. It cannot see the driver, and it quotes only the wait the shop itself sets, so it has no live estimate to offer and will not invent one. Those calls need a person with a dispatch screen.

And if you are running five brands out of one kitchen on one number because that number is on every listing, the fix is the numbers, not the voice. Split the lines, then decide how many of them deserve an agent. If one of your brands is pizza, the pizzeria guide goes deeper on the half-and-half problem.

FAQ

Final Thoughts

The decision at a ghost kitchen is not whether a voice agent can take an order. It is whether the line has a brand attached to it, and whether the caller has anywhere to pick the food up.

Get those two right and the phone becomes a commission-free channel instead of a number nobody answers. Get them wrong and you have added a fourth place for a wrong order to happen.

Kovo answers one brand's line, reads the order back, takes payment on the line or by text link when cards are on, and prints the ticket to the kitchen, from $99 a month per store. Fourteen days, no card, and a number for each brand. Take a test call.