Four conversations, end to end
Yumoto Kagetsu is a 24-room onsen ryokan. Pick a scenario on the left; tap the quick replies and card buttons as a guest would. The right-hand pane is the property’s view of the same conversation — four KPI tiles and the event log.
Type something off-script and the bot will tell you it would rather not guess, and hand the thread to a person with a summary. That is the real behaviour.
What each conversation is meant to prove
Party, plan and room type are already known, so the whole stay is two taps.
12 September, 19:00. Mr Tanaka stayed on 3 and 4 November last year, in Kaede, on the kaiseki plan. He booked that stay through an OTA.
One sentence becomes a room type, a plan, a child price and a held stay.
22:10. A guest we have never met sends the whole enquiry in a single message. The front desk closed two hours ago.
The concierge answers in the guest's language and works back from the kitchen.
16:20 on the day of arrival. An inbound guest with a delayed flight writes in Chinese, four hours before the kitchen's cut-off.
The desk closed at 22:00. The request still reaches a person, and comes back.
23:04. A guest in Kaede messages from the room. The front desk is unstaffed and the night porter is on the floor.
The same four scenarios drive the compact demo on the product page and the industry page for hotels and ryokan. They are held in /hotel-concierge/demo.json, and the builder’s simulator replays them against your own plans and prices once you have imported them.
Run it against your own plans next.
Import your room types and rate table and the same four conversations replay with your prices in them.