WhatsApp AI Assistant for Hotels: 24/7 support, 90% deflection, +35% upsell

Context
In hospitality and outdoor hospitality, front-desk teams handle WhatsApp messages around the clock: late check‑in questions, amenity requests, urgent incidents, and upsell opportunities. Keerok built a dedicated conversational AI assistant to standardize service quality, slash response times, and capture incremental revenue, while integrating cleanly with each property’s existing tools.

Challenge
We had to deliver instant, multilingual replies; ground answers in reliable data (rates, policies, amenities) spread across sources; correctly route sensitive intents (emergency, health/safety, complaints); and avoid hallucinations. We also needed performance tracing, accurate routing to the right teams, and a scalable multi‑property rollout without diluting each brand’s tone.

Solution
We engineered a platform on Django 6 + DRF, orchestrated by Celery/Redis. The conversational core combines LangChain and Mistral AI with hotel‑versioned prompts and LangSmith tracing. A dual‑path intent engine (rules/regex + LLM) covers dozens of intents with automatic workflow routing (emergency, upsell, booking…). A PostgreSQL + pgvector RAG ingests website, PDF and DOCX, builds Mistral embeddings, and serves FR/EN semantic search. End‑to‑end WhatsApp automation (Meta Cloud) handles classification, retrieval, generation, and rich delivery (interactive buttons, location cards). Integrations include Mews, GitHub Actions CI/CD, and Docker.

Results
<5s first response time, 90% of requests resolved/deflected by AI on WhatsApp, and +35% upsell on identified opportunities. The assistant runs 24/7 and is deployed across over 15 properties, freeing staff for high‑value tasks and sensitive cases. Real‑time alerts and targeted escalation secure critical situations, while per‑hotel prompt versioning preserves each brand’s voice.