Case Study: Saving 20 Hours a Week for a Real Estate Agency with AI Agents

Real Estate is a numbers game. But it’s also an exhaustion game.
​I recently consulted for a mid-sized agency here in Lagos. Their problem wasn't a lack of leads; it was Lead Fatigue.
For every 100 people who messaged them on WhatsApp asking "How much?", only about 3 were serious buyers with the budget to proceed.
​The agents were spending 4 hours a day just answering the same three questions:
​"Is it still available?"
​"What is the price?"
​"Can I see pictures?"
​They were human FAQs.
​I proposed a solution: Let’s fire the humans from the 'First Response' layer and hire an AI Agent.
​Here is how I built a system that not only answers questions but qualifies leads and books inspections automatically.
​The Architecture
​We didn't want a "dumb" chatbot that just gives static replies. We needed an Agent that could query their specific database of properties.
​The Stack:
​Brain: OpenAI (GPT-4o) via LangChain.
​Communication: Twilio (WhatsApp API).
​Database: Supabase (PostgreSQL + pgvector).

​Knowledge Base: A live sync of their property listings.
​Phase 1: The "RAG" Knowledge Base
​The biggest challenge was accuracy. The AI couldn't hallucinate a price.
We used RAG (Retrieval-Augmented Generation).
​When a user asks, "Do you have any 3-bedroom flats in Ikeja under N5m?"
​The system converts the query into a vector.
​It searches Supabase for matching properties.
​It retrieves the exact data (Price, Location, Features).
​It feeds this to GPT-4o to generate a polite, human-like response.
​Result: The bot never guesses. It only sells what is in stock.
​Phase 2: The "Guardrail" Qualification
​We instructed the AI to act like a senior sales agent. Its goal wasn't just to chat; it was to qualify.
​System Prompt Snippet:
​"You are a helpful Real Estate Assistant. Your goal is to get the user to book an inspection. Before booking, you must politely ask for their budget and timeline. If they cannot afford the property, politely suggest cheaper alternatives."
​This filter alone saved the human agents hours of driving to inspections with clients who had zero intention of buying.
​Phase 3: The Handoff
​If the user is serious and agrees to a time, the AI Agent uses a "Tool" (via LangChain) to check the human agent's Google Calendar and book the slot.
The human agent gets a notification: "New Inspection Booked: Mr. Obi, 3 Bedroom, Budget Verified."
​The Results
​After 30 days of running this pilot:
​Response Time: Dropped from ~2 hours to <1 minute.
​Agent Workload: Reduced by ~20 hours/week (no more answering "How much?" at 10 PM).
​Conversion: Inspection bookings increased by 15% because the bot replied instantly while leads were hot.
​Conclusion
​We are past the era of "Chatbots." We are in the era of "AI Agents."
A chatbot follows a script. An Agent uses tools, makes decisions, and performs work.
For this agency, it was the difference between being busy and being profitable.

​Hi, I'm Frank Oge. I build high-performance software and write about the tech that powers it. If you enjoyed this, check out more of my work at frankoge.com