AI Concierge for Hotel Guest Planning and Booking

The challenge

OASIS212 had no chatbot. Visitors with real questions before booking had two options: dig through the site, or email/whatsapp reception staff and wait. I tested a keyword-matching bot first. It failed almost immediately: during initial test if you asked it about bringing camera gear you get shown kitesurf gear rental prices, because the bot matched on the word “gear” and stopped listening there. That confirmed the actual problem: a system built to match words, not understand people.

The approach

Guests don’t ask questions in neat categories, they ask the way they’d ask a person standing in front of them. I built the assistant to follow actual intent instead of trigger words. Supporting elements, room photos, the booking form, a way to leave a phone number, stay available but only surface when they’re relevant to what’s actually being asked. The harder problem was restraint. Left unconstrained, an AI answers everything at once and quotes prices nobody asked for yet, which reads as pushy. I built it to answer what was asked, offer more only when useful, and hold off on selling before the guest has said hello.

The execution

Tools used: Claude Code

1. 01Gave the assistant a name and consistent personality, Meryem, instead of a generic chat widget, so guests get one voice from their first question through check-in
2. 02Rebuilt booking logic around actual guest behavior: capture the stay booking while interest is high, resolve smaller requests like lessons or spa treatments once the guest is already committed
3. 03Every guest conversation is logged and passed to the team with full context attached, what the guest asked, what they were told, what they need, so whoever follows up picks up where Meryem left off instead of asking the guest to re-explain themselves from scratch
4. 04Tuned response length and pacing so answers don’t over-explain policies nobody asked about or dump text on simple questions
5. 05Built an analytics dashboard tracking what guests ask, where conversations drop off, session duration, and response times, to measure guest intent and keep improving the user experience