Conversational AI
A post-booking concierge that knows the trip, and knows when to hand over.
A WhatsApp agent that picks up after a guest books, helps plan the rest of the trip from the marketplace’s own catalogue, and passes real support actions to people.
The problem
Where the work was stuck.
After a guest booked one experience, the company went quiet until the next ticket, while the traveller planned the rest of the trip across search, social and chat assistants. The days before arrival are the highest-intent part of the journey, and email, the only line in, drew little engagement.
What we built
A live concierge running across several cities.
A WhatsApp-native concierge that opens with the guest’s real booking, helps plan the rest of the trip, recommends from the marketplace’s own experiences, and hands anything it should not do itself to a person.
How we worked
How the work took shape.
We designed the conversation and the agent around one rule: each reply has one job. A memory step holds the trusted trip details, a director picks the single objective for the turn, retrieval draws on the catalogue first, one writer drafts the reply, and a validator checks it is grounded before anything is sent. Questions about a booking are answered from the booking itself; cancellations, refunds and changes go to a human. Openers, recommendation formats and offers ran as live experiments, judged on the quality of replies, not the reply rate alone.
Your next step
Where does your work get stuck?
Bring us the process. We’ll help you work out what it needs.
Bring us a problem