❓ Problem
Picture a fully booked Friday in Manhattan. You've ordered the turbot, rostered two extra servers and turned away walk-ins. Then a chunk of the room doesn't show, because a chunk of the room was never really coming.
No-shows were already a slow bleed for American restaurants. OpenTable's platform has historically averaged 5-7% no-shows on a typical night, with operators reporting rates above 20% on holiday weekends when no deposit is required. At an 8% baseline, a 120-seat upscale-casual spot taking 900 reservations a week eats roughly 72 no-shows every week. Diners holding tables at three hot spots and deciding at 7pm has become so normal that operators are now calling it a no-show "epidemic".
Now add agents. This month a New York VC had his Resy account deactivated after telling an AI assistant from Instinct to land him a table at 4 Charles Prime Rib. And this is before the category goes mainstream: Instinct recently raised $350m, and Meta's Muse agent has already passed 2 million downloads.
The kicker is that a restaurant can't tell who's on the other end. It might be a real guest's assistant, a bot hoarding tables for resale, or someone deliberately trashing your Friday. The resale economy already exists: Dorsia charges up to $25,000 a year for access, and gray-market sites like AppointmentTrader let deep-pocketed diners skip the line. Agents make hoarding cheaper to run at scale. The booking system says "confirmed". The empty table says otherwise.
✅ Solution
An AI bouncer that sits between the booking system and the restaurant's inventory, starting with restaurants.
- Score every booking on arrival. Speed of the booking flow, repeated attempts across slots, account history, card behaviour and whether the request carries a verified agent identity. Each reservation gets a risk rating before it eats a table.
- Wave the regulars straight in. Low-risk bookings go through untouched, so real guests never notice it exists.
- Challenge the suspicious ones. Ask for an extra confirmation, a card hold or a deposit, cap how many future reservations one identity can hold, and auto-release unconfirmed tables back into inventory ahead of service.
- Let good agents in on the restaurant's terms. Verified assistants can book within rules the venue sets: max party size, max bookings per week, confirmation required by 4pm on the day.
- Show the owner the receipts. A dashboard of which bookings were challenged, which confirmed, which walked, and how the no-show rate moved.
📊 Key Numbers
Market size
- There are 154,196 single-location full-service restaurants in the US in 2026, part of an independent full-service segment worth $275.7bn this year. At ~$99/mo, US independents alone are a ~$180m annual software opportunity (TAM), before chains or other verticals.
- The distribution prize: OpenTable helps 60,000+ restaurants fill 1.8 billion seats a year. Charge a platform two cents per protected cover across that volume and it's a ~$36m line from one partner.
- Restaurants already pay per cover for bookings: a venue doing 1,500 network covers a month pays OpenTable $1,500-2,250 in cover fees alone. A protection layer priced at a fraction of that is an easy line item to justify.
- The adjacent category is real but estimates vary wildly. One report puts restaurant reservation software at $1.2bn in 2025, with North America accounting for $890m of it. Treat it as directional.