Gurugram multi-brand restaurant group, 9 outlets, POS with an API, WhatsApp API through a BSP, catering enquiries by phone
Week 1Intent mapClassify 1,000 past chats; rule of thumb: buttons where an intent has under 10 fixed answers, free text with an LLM otherwiseMenu, booking, catering and feedback flows; complaints go to people
Weeks 2–3Build flowsWhatsApp Flows for table booking, list messages for outlets (10 rows max, so group by area), a small LLM layer answering menu questions from approved dataBot running on a test number
Week 3Handover and trackingShared-inbox handover with chat context; drop-off events logged per stepFunnel by step visible
Weeks 4–5Launch and tuneGo live, review drop-offs weekly; most common drop-off: typed date or address questions where a Flow picker would doFlows simplified at the worst drop-off points
Where it starts1,500 chats/week, 70% about menus, timings and table bookings, 20% of chats unanswered after closing time
Where it endsThe group answers routine chats round the clock with clean handovers. Weekly they watch bookings via the bot, drop-off by step and handover rate.