The actual problem
A full-service restaurant in San Diego running dinner service has one or two people at the front. When it gets busy — 6:30 on a Friday, every table turning — those people are seating guests, managing the waitlist, and keeping the floor moving. The phone rings. It goes to voicemail. The caller books somewhere else or just shows up somewhere they didn't plan to.
During off-hours it's different: the phone rings at 2pm with someone asking about hours, whether you're open on July 4th, or wanting to make a reservation for Saturday. A person answers, but they're also doing side work, helping prep, or just not in a position to give that call full attention.
Voice AI doesn't solve all of this. But it covers a specific slice of it well: inbound calls that need a consistent, immediate answer about hours, availability, or reservations.
What a voice AI system actually does
A well-built restaurant voice AI handles four things reliably:
- Hours and location questions. "Are you open Sunday brunch?" "Where are you located?" "Do you have parking?" These are entirely answerable by a voice agent with your knowledge base loaded.
- Reservation intake. For restaurants using OpenTable, Resy, or SevenRooms, the AI can check availability and hold or log the request. This requires an integration — it's not automatic — but it's straightforward to build.
- Menu questions at a basic level. Common allergen questions ("is the pasta gluten-free?"), general dish descriptions, whether you have vegetarian options. Specific modification questions or detailed prep questions should escalate to a human.
- Escalation on anything it can't handle. A caller wants to place a large private-event inquiry, or they're calling about a past visit issue — the system should recognize this and either warm-transfer to staff or take a message for callback. This is one of the harder parts to get right.
What it doesn't do well: highly specific table requests, complaints or conflict resolution, anything that requires emotional attunement, and calls in languages you haven't configured it to handle. Knowing the limits matters as much as knowing the capabilities.
The build: what you're actually paying for
The infrastructure for a restaurant voice AI runs on three pieces: Twilio (your phone number and call routing), Vapi (the voice AI layer that manages the conversation), and OpenAI (the language model that understands and responds). All three have published pricing. There are no hidden platform fees — you pay them directly after we build it.
What you're paying a builder like us for:
- System design and configuration. Prompt engineering, conversation flow design, fallback logic. This is where most builds fail — underfunded prompt work produces a bot that sounds like a bot and frustrates callers.
- Integration with your reservation system. Toast, Square, OpenTable, Resy — these each have their own APIs and quirks. A generic voice AI off the shelf won't be connected to your actual data.
- Testing and voice tuning. Real-call testing with different phrasings, accents, fast talkers. The system needs to handle the range of people who actually call San Diego restaurants.
- Documentation and handoff. You need to know how to update your knowledge base when your hours change, how to add seasonal menu items, and what to do when something breaks.
Our restaurant voice AI pilot starts at $2,500 for setup, with a $400/month managed option if you want us handling ongoing maintenance, updates, and monitoring. After that, your ongoing infrastructure cost to Twilio and OpenAI is typically $30–$50 per month for a full-service restaurant — you pay that directly to them.
On competitor pricing
Subscription-based voice AI services (Hostie, Slang, others) charge $199–$599/month per location, running on their infrastructure. That works for some operators. The tradeoff: when you stop paying, the system stops. The build cost we charge covers something you keep — on your Twilio account, your OpenAI keys. We think that matters, especially once the system is working and you want to modify it over time.
Is it worth it for your restaurant?
The math works when calls are a genuine friction point — a busy dinner house with hosts stretched thin, or a takeout-heavy operation where the phone rings constantly. It doesn't make sense if your staff has bandwidth, or if your call volume is low enough that a missed call is genuinely rare.
The honest question to ask: how many calls do you miss in a typical week during service? If the answer is "some, consistently" — and if those calls have any chance of being reservations — the math on a $2,500 build pays off quickly. A single no-show-prevented reservation covers a meaningful fraction of it.
If you're not sure, the Operations Audit we offer ($1,500) will tell you whether voice AI makes sense before you spend on it. We map your actual call patterns, figure out what your phone is handling and when, and give you an honest recommendation — including if the answer is to handle it a different way.
What to do next
If this sounds like something your restaurant needs, the right starting point is a conversation — not a commitment. We spend 30 minutes understanding your setup, your current phone handling, and your reservation system, and we give you a clear picture of what's involved and whether it pencils out for you.
No pitch deck, no discovery call designed to end in a proposal. Just the information you need to make a decision.