• AWS-native AI integration · ships in 6–10 weeks

Restaurant group answers 99% of reservation calls

  • Voice AI · Hospitality
  • Two-location upscale restaurant group — ~250 covers/night, Toast POS + OpenTable

The problem

Front-of-house was answering phones during service — bad for the guest in front of them and the guest on the phone. The GM estimated 15–20% of inbound reservation calls during peak service went unanswered; modifications were a particular drain.

Our approach

What we built

  • Inbound voice agent handling new reservations (with OpenTable writeback), modifications and cancellations, hours/menu questions, allergy and dietary notes, and large-party callback capture.
  • Bilingual English / Spanish routing detected on the first turn.
  • Hard rule: the agent never quotes wait times for walk-ins — that’s a service-floor call. It captures the callback request and pages the host stand.
  • Daily transcript digest emailed to the GM at 11am next-day so service-floor issues get caught fast.

Caveats

Mispronouncing chef-driven menu names in either language annoys real customers; we hand-tune pronunciation for the top 30 dishes per location. The agent was occasionally too eager to upsell tasting menus; we removed that behavior in week 3 after a guest complaint. POS integration is real work — Toast, Square, Clover, and Qu all need their own connector.

Stack. Vapi · AWS Bedrock Claude · OpenTable API · Toast read-only for hours and menu lookups · bilingual routing.

Outcome

Reservation calls answered moved from ~80% to ~99%. Staff time on phone reservations per shift fell from 2 hours to ~15 minutes. 14% of weekly reservations now come in after hours.

99%
reservation calls answered (was 80%)
87%
reduction in phone time per shift
14%
weekly reservations from after-hours