The Situation
The brokerage ran almost its entire top-of-funnel on WhatsApp. Buyers and renters messaged at all hours asking the same questions — "Is this listing still available?", "What's the price per square foot?", "Can I see it Saturday?" — and a team of 40+ agents tried to keep up between showings, calls, and paperwork.
The math didn't work. Messages that landed in the evening or over the weekend often sat unanswered until the next business morning, by which point the prospect had already messaged three other brokerages. Agents were copy-pasting the same listing details dozens of times a day, and high-intent buyers were buried in the same inbox as tire-kickers. Leadership estimated that a meaningful share of inbound demand was leaking out simply because no one replied fast enough.
They didn't want a generic chatbot that frustrated serious buyers. They wanted something that actually knew the inventory, could hold a real conversation, book a viewing, and hand off cleanly to a human the moment a deal looked live.
What We Did
We built a WhatsApp-first AI concierge backed by a retrieval-augmented (RAG) layer over the brokerage's live listing database. Every inbound message is matched against current inventory, so the assistant answers with accurate, up-to-date details — price, availability, location, amenities — instead of hallucinating.
The build had four moving parts:
- RAG over listings. Listings sync from the CRM into a vector store, refreshed on every update. The assistant grounds every answer in real records and never quotes a unit that's already off the market.
- Conversational booking flow. When a prospect shows buying intent, the concierge collects the essentials, offers real viewing slots from agent calendars, and confirms the appointment — all inside WhatsApp.
- Smart escalation. A lightweight intent classifier watches for signals like budget confirmation, financing questions, or repeated follow-ups, then routes the conversation to the right human agent with the full transcript attached.
- CRM write-back. Every conversation, qualification detail, and booked viewing is written back to the CRM, so agents pick up exactly where the AI left off.
We ran the whole thing in a shadow mode for the first week — the AI drafted replies, humans approved them — to tune tone and accuracy before letting it respond on its own.
The Results
Within six weeks the concierge was resolving 62% of inbound inquiries end-to-end with no human involvement, and first-response time dropped from several hours to about three minutes — at any hour of the day. Because the assistant booked viewings directly into agent calendars, monthly viewings climbed roughly 38% without adding headcount.
Just as importantly, agents stopped drowning in repetitive questions and spent their time on the conversations that actually move deals forward. The brokerage now treats the concierge as the always-on front desk for its entire pipeline.
