Real estate agents at Maplecrest Realty were missing inbound calls constantly — during property showings, during negotiations, and after office hours. A missed call in real estate is rarely a voicemail. Buyers move fast and call the next agent.
Even when agents were at their desk, a significant portion of inbound call volume was low-complexity scheduling enquiries: 'Is this property still available?', 'Can I book a viewing this Saturday?', 'What times does the agent have free?' These calls demanded availability, not expertise — and they were consuming agent time that should have been spent on active buyers and deal negotiations.
The agency needed a solution that could answer every call, sound professional, qualify the prospect, and book the viewing without the agent ever picking up the phone.
I built Stevenson — a Vapi voice agent with a GPT-4o language model, connected to n8n for backend logic, Google Calendar for real-time availability, Twilio for SMS confirmations, and Google Sheets for lead tracking. Stevenson answers every call, regardless of when it comes in.
Stevenson is configured with a custom system prompt, tuned end-of-utterance sensitivity for the natural pauses in real estate conversations, and interruption handling so callers can speak naturally rather than waiting for the agent to finish. The GPT-4o language model powers the conversation, and the voice profile is set to sound professional and unhurried.
The agent opens with a professional greeting and moves into a structured qualification flow — collecting budget range, preferred location, property type (residential or commercial), bedroom count, and purchase or rental timeline. The conversation branches based on responses, so a first-time buyer follows a different flow from an investor looking for commercial space. All collected data is structured for downstream calendar and CRM use.
Using a Vapi tool call that hits an n8n webhook, Stevenson queries the agent's Google Calendar in real time. Available slots are returned to the conversation and offered to the caller in natural language — 'I have Tuesday at 2 PM or Thursday morning at 10, which works better for you?' — without reading raw timestamps or requiring the caller to navigate a booking page.
The selected slot is created in Google Calendar via the n8n webhook with a structured event description containing all qualification data. Calendar conflict detection prevents double bookings — if a slot becomes unavailable between the availability check and the booking confirmation, the agent offers the next available option without breaking the conversation flow.
Within 30 seconds of the call ending, Twilio sends the caller an SMS confirmation with the viewing date, time, property address, and the agent's name. Simultaneously, a lead record is written to Google Sheets with the caller's full qualification data — name, phone, budget, preferred area, property type, timeline, and the booked appointment details — giving the agent a complete brief before they arrive at the viewing.