AI voice agent for restaurants: what it handles, what it costs and how to choose one

The short answer

An AI voice agent for restaurants is software that answers your phone in a natural voice, takes pickup orders and reservations, answers questions about hours and the menu, and transfers anything unusual to a person. It pays off when calls go unanswered during the dinner rush, when the host stand is busy seating guests, and after closing.

In 2026 there are three ways to get one: a restaurant-specific product such as Yelp Host or Slang.ai, a voice feature or integration offered through your POS, or a custom voice agent built on a platform like Retell AI and connected to your POS, reservation system and reports. If an off-the-shelf product supports your POS and reservation system and follows your menu rules, start there. Custom builds make sense for multi-location groups, complex menus, catering and private events, or when you want the call data in your own accounts.

What restaurant callers actually ask for

Before you compare vendors, look at what your calls are. The most detailed public data we found is Maple's restaurant phone audit, published in spring 2026 from calls handled for its paying restaurant customers. Maple notes that the sample is its own merchants, not all US restaurants, but the pattern is useful:

• 19% of calls arrived outside business hours. Those calls go to voicemail or nowhere unless something answers them.
• The peak hour was 5 PM local time, right as dinner service starts, and Friday carried 20.1% of weekly calls against 11.2% on Monday.
• At restaurants that take orders, 39.8% of calls were to place a pickup or delivery order, 13.6% asked to speak to a person, 7.0% were reservations, 4.3% asked about hours and 2.3% about the menu or prices.
• At reservation-led restaurants, 28.3% of calls were bookings, 14.2% wanted a person and 11.6% were orders.

Phone ordering has not gone away either. A PYMNTS and Paytronix study from October 2022, based on a survey of more than 2,200 US adults, found 42% had ordered from a restaurant by phone in the previous three months.

The takeaway: the busiest phone hour lands exactly when your staff are least able to answer, and most calls are routine. That is the job an AI voice agent is built for.

What an AI voice agent should handle, call by call

Write down how the agent should treat each call type before you buy. Most failures come from vague rules, not from the voice technology.

Call type	What the AI should do	Hand to a person when
Pickup order	Take items with modifiers, check items that are 86'd, read the order back, quote total and pickup time, send to POS	Large orders over a limit you set, or the caller insists on a substitution the menu does not allow
Delivery order	Send the caller your delivery link or take the order only if your POS handles delivery zones and fees	Address outside your zone
Reservation	Check live availability, book, take party size and notes, text a confirmation	Parties above your large-party threshold, or special occasions needing a set menu
Waitlist and wait time	Quote the current wait from your waitlist system and add the guest	Rarely
Hours, parking, dress code, menu questions	Answer from an approved fact sheet	Questions not on the sheet
Allergy questions	Share only what you have approved in writing, and offer a staff callback	Any severe allergy or cross-contact question
Catering and private events	Collect date, headcount, budget and contact details and log the lead	Always for quoting
Complaint or refund	Apologize, take details, transfer or promise a manager callback	Always
Vendors and sales calls	Take a message or end the call politely	Never

Notice the 13.6% to 14.2% of callers in Maple's data who asked for a person. A good agent does not fight them. It offers to help, and if the caller still wants a person it transfers or takes a callback with a summary.

The 2026 options compared: Yelp Host, Slang.ai and custom builds

Here is what the main options publish, checked in October 2026. Prices change, so confirm terms with each vendor.

• Yelp Host. Yelp's press release of 28 July 2026 says Host has handled more than 1 million calls since its October 2025 launch. The same update added food ordering for pickup at counter-service restaurants, an OpenTable integration for the US and Canada, 16 new languages with automatic language detection, new voices and call routing to different team members. Yelp says Host connects to Toast, Square and Clover through Stream. Price: from $249 a month, or $199 a month for Yelp Guest Manager customers.
• Slang.ai. Slang's pricing page lists Core at $399 a month per location, with reservation management, text confirmations, call recordings with summaries and reservation integrations, and Premium at $599 a month per location, adding support for 97 languages, bilingual support, custom branding and reservation cross-selling. Enterprise is priced by quote. Slang is built mainly around reservations and guest questions, so check how it handles your ordering flow before you commit.
• Custom voice agent on a voice platform. Platforms such as Retell AI sell the voice layer by the minute. Retell's pay-as-you-go pricing is $0.07 to $0.31 a minute, depending on the voice and AI model you choose, and its restaurant page lists integrations with Toast, Square, Clover, OpenTable and Resy. You (or an agency) write the call flows, connect your systems and own the setup.
• POS-side options. Some voice ordering vendors integrate directly with POS systems. Toast's support site, for example, documents an integration with InceptAI. Check your POS marketplace first, since an approved integration is usually the most reliable way to get orders onto the kitchen display.

Question	Restaurant product (Yelp Host, Slang.ai)	Custom voice agent
Time to go live	Days. Slang says setup takes under 30 minutes	Typically 3 to 8 weeks in stages
Phone ordering into POS	Depends on the product and your POS; confirm modifiers and 86'd items	Yes, if your POS has an API or approved integration
Custom rules (catering, large parties, set menus)	Limited to vendor settings	Full control
Several locations or concepts	Priced per location	One agent can route by location and concept
Who owns scripts, recordings and number	The vendor	You, in your own accounts
Best for	Single restaurants on a supported POS	Groups, ghost kitchens, catering-heavy and franchise operators

What an AI voice agent for restaurants costs: a worked example

Costs come in three parts: a subscription or platform fee, usage (minutes, texts, phone number) and any setup work to connect your POS, reservations and reporting.

Take an illustrative restaurant that sends 900 calls a month to the AI, averaging 2 minutes each. That is 1,800 minutes. Use your own call logs, because volume varies widely.

Option	Published pricing	Monthly cost for one location
Yelp Host	From $249 a month ($199 for Guest Manager customers)	From $249; ask how minutes and ordering are billed
Slang.ai Core	$399 a month per location	$399
Slang.ai Premium	$599 a month per location	$599
Custom agent on Retell AI	$0.07 to $0.31 a minute	$126 to $558 in usage, plus phone number and a one-time build fee

For a single restaurant, the off-the-shelf products are usually the cheaper start. A custom agent costs less per month once usage is the only bill, but the build is the larger cost. Our guide to AI automation agency costs explains how those build fees are priced. At ten locations, per-location subscriptions multiply by ten while one custom agent serves all of them, which is where the maths often flips.

Work out your own return. Pull last month's missed and abandoned calls from your phone provider. Estimate what share were orders (Maple's data suggests roughly 4 in 10 at order-taking restaurants, but use your own). Multiply recovered orders by your average phone ticket from your POS reports. If your average phone order were $40 and the agent captured 40 extra orders a month, that is $1,600 in sales. Compare that against the monthly fee, and also against the cost of adding a host or phone shift at your local wage.

How to roll out an AI voice agent at your restaurant, step by step

1. Pull 60 to 90 days of call data. Calls by hour and day, missed calls, voicemails and average call length. Find the windows where calls go unanswered.
2. Clean up the menu in your POS. The agent reads what is there. Fix item names guests actually say, modifier groups, required choices, prices and what is available at lunch versus dinner.
3. Write your rules. Order limits, pickup lead times by hour, large-party thresholds, deposit rules, delivery zones, allergy wording and who gets transferred calls on each shift.
4. Decide the routing. Common patterns are after hours only, overflow after a set number of rings, or AI first with a transfer option. Start with after hours and overflow.
5. Connect systems. POS and kitchen display for orders, OpenTable, Resy or your table system for bookings, and a dashboard or sheet for every call outcome.
6. Test with real calls. Replay 40 to 60 real orders and bookings, including messy ones: half-and-half pizzas, "no onions, extra sauce", a caller who changes their mind, a party of 14, and a caller switching languages.
7. Go live in shadow mode. For one to two weeks, a manager checks every AI order ticket and booking before service, and listens to a sample of calls daily.
8. Review weekly. Track answered, ordered, booked, transferred and abandoned calls by hour, and fix the scripts where calls fail.

Website and Instagram questions can be handled by an AI chatbot that shares the same fact sheet, and catering leads can flow into your CRM with workflow automation.

Payments, recording and outbound call rules

A restaurant voice agent records calls, may take payments and may send texts, so a few rules apply. This is practical guidance, not legal advice. Ask your attorney to review your scripts.

• Card payments. Do not let card numbers be read aloud into a recording or transcript. Use a secure payment link texted to the caller, pay at pickup, or a payment flow provided by your POS or payment processor that keeps you within PCI DSS requirements.
• Call recording consent. Some states, including California, require every party to agree to a recording. A short notice in the greeting is the common approach.
• Tell callers it is automated. A one-line disclosure sets expectations and makes it easy to ask for a person.
• Outbound calls are different. On 8 February 2024 the FCC ruled that AI-generated voices count as "artificial" under the TCPA. Calling guests to confirm reservations or fill cancellations with an AI voice needs the consent that applies to prerecorded calls.
• Texts need consent. An order confirmation the guest asked for is different from a promotional text. Keep marketing separate and honor STOP replies.

Questions to ask before you choose a restaurant voice agent

• Does it take orders into my POS, or only answer questions and send links?
• Does it read modifiers, required choices and 86'd items in real time?
• Which reservation systems does it book into directly?
• How does it take payment, and are card details ever stored in transcripts?
• What does my busiest month cost, including minutes, texts and per-location fees?
• Can it route by location, and to different staff by shift?
• Which languages does it handle, and can I hear a test call in each?
• Can I export every recording, transcript and outcome?
• Who owns the phone number and scripts if I cancel?

For questions to ask any automation partner, see how to choose an AI automation agency.

Common mistakes, and when not to buy

• A messy POS menu. If the menu has items called "SPCL 2" and modifiers nobody uses, the agent will read them out or take wrong orders. Clean it first.
• No 86'd sync. An agent that sells the last special twice creates the angriest customers of the night.
• Letting the AI answer allergy safety questions freely. Limit it to approved statements and offer a staff callback for anything serious.
• Judging on the demo voice. A natural voice matters less than whether tickets land correctly on the kitchen display.
• Nobody reads the transfers and hang-ups. Every failed call is a script fix waiting to happen.

Skip it or wait if: you already answer nearly every call; almost all orders come through online ordering and delivery apps; your POS has no integration or API for orders, so the agent could only take messages; or your problem is no-shows rather than missed calls, in which case reminders and deposits pay back faster.

Getting an AI voice agent built for your restaurant

If a restaurant product supports your POS and reservation system and follows your rules, use it. Bring in an agency when you run several locations or concepts, take catering and private events by phone, use systems the products do not support, or want call outcomes reported next to sales.

We build AI voice agents and AI receptionists for service businesses. Every project starts with a free review of your call data, gets a fixed price, ships in stages and runs on accounts you own, including the phone number, scripts and code.

To get started, book a free review and tell us which POS and reservation system you use and roughly how many calls you miss on a Friday night.

FAQ

Frequently asked questions