Restaurants are the niche every new voice AI operator tries first and the one that quietly kills more agencies than any other. The logic looks airtight — restaurants have phones, the phones ring, nobody answers during a rush. So you build a reservation-and-ordering agent, pitch fifty spots, and discover the brutal truth: most restaurants are a terrible voice AI business. This is the honest map of when to walk away and the narrow slice where it actually prints.
Why most restaurants are a trap
- The budget isn't there. An independent restaurant runs on 3–5% net margins. A $500/month tool is a real line item they will scrutinize forever. Compare that to a dental practice or insurance agency where $1,200/month disappears into the noise. Same build, one-fifth the willingness to pay.
- The value is capped and seasonal. A missed reservation is worth maybe $60–$120 in covers. A missed insurance quote is worth a policy with lifetime value in the thousands. The math that makes voice AI obvious in other niches barely clears the bar here.
- They already have Toast, Resy, OpenTable. Reservations and online ordering are largely solved by tools the restaurant already pays for. You're not filling a gaping hole; you're shaving a thin sliver.
- Menus change constantly. Specials, 86'd items, seasonal swaps, price updates. Every change is a knowledge-base update. A niche where the knowledge base drifts weekly is a support burden that eats your margin alive.
- Turnover kills your champion. The manager who signed loves you. They leave in four months — restaurant turnover is savage — and the new one has never heard of you. Churn you didn't cause.
If you're pitching a single neighborhood bistro on a $400/month reservation bot, stop. You've picked the hardest possible version of this business.
When it actually works
The slice that pays has three traits: high call volume, real revenue per call, and multi-location budget.
- Multi-location groups and franchises. Ten locations under one owner changes everything. Now you're selling one integration across ten phone numbers, the budget is regional not per-store, and the buyer thinks in systems. $2,000–$4,000/month for a group is easy where $400 for one store is impossible.
- High-ticket / high-volume formats. Pizza chains and delivery-heavy concepts where the phone is the revenue channel — hundreds of orders a night, real dollars per call, and a genuine staffing problem during the rush. Order-taking with upsells ("garlic knots with that?") has measurable lift you can put on a slide.
- Catering and private events. This is the hidden gem. A catering inquiry is worth $800–$5,000, comes in by phone, and routinely hits voicemail because the line staff can't handle it. An agent that captures the event details — date, headcount, budget, contact — and routes a warm lead to the events manager is pure margin protection. Sell that, not "we answer your phone."
The wedge use-case
Lead with catering/event lead capture + after-hours order-taking, never generic reception. The pitch is: "How many catering inquiries went to voicemail last month, and what's each one worth?" That reframes you from a cost (a $400 phone bot) to a revenue tool (capturing $2,000 events that were leaking).
What to charge
For a group: $1,500–$3,500/month plus a $1,500–$3,000 setup, priced per location or per captured lead. Vapi or Retell for voice, POS/reservation integration where it exists (Toast, Square, Resy APIs), warm transfer for anything complex. Skip single independents unless they're a lighthouse reference you're taking at a loss.
The honest caveat
Two hard limits. Payment capture over the phone drags in PCI scope you don't want — route paid orders to a link or a human, don't read card numbers to the agent. And order accuracy is unforgiving — a wrong pizza order is an angry customer and a refund, so keep the menu tightly scoped, confirm every item back, and 86 anything the kitchen flags in real time. Respect those two lines and the catering-plus-groups slice is a real business. Ignore the map and chase corner cafés, and restaurants will teach you why operators warn against them.