CAIDEN — Booked for Real Estate

The problem

Real estate agents spend 30–50% of their day on phone calls — listing inquiries, viewing requests, qualification questions, follow-ups. Every missed call is a potential commission lost, and buyers looking for a home rarely call during business hours. They call from the parking lot after a showing, at 9pm when kids are in bed, on Sunday mornings while browsing Zillow. Brokerages tried IVR systems and offshore answering services and burned trust with both. IVRs frustrate hot buyers into hanging up. Offshore agents lack local market knowledge — they can't answer "what school district?" or "is that neighborhood walkable to the train?" — and callers can tell within one exchange. The market needed an AI that could handle the conversation naturally, qualify the lead, understand the local property landscape, and route to the specialist agent.

Key challenges

Real estate has vocabulary and context problems that generic voice AI fails at. Callers reference addresses ambiguously ("the one on Oak"), use MLS jargon inconsistently ("townhome" vs "row house"), and expect the receptionist to know current price cuts, days on market, and comparable listings. Qualification requires nuance — a first-time buyer with a pre-approval letter and a seasoned investor need different handoffs to different agents. And the calendar problem is real: agents' calendars are hostile terrain of double-bookings, blocked time, and location constraints.

What we built

Booked is CAIDEN's real-estate vertical — the same core platform (LiveKit, Deepgram, ElevenLabs, Fly.io) with real-estate-specific conversation flows, MLS data integration, and agent-routing logic tuned to property type, price range, and neighborhood specialization. Call flow: pickup, greet with the brokerage's branded voice, listen for listing reference, disambiguate against live MLS data, answer neighborhood/school/HOA questions from a per-listing knowledge base, qualify budget and timeline, propose viewing slots from the appropriate agent's live calendar, book with double-verification via SMS confirmation, and route follow-up questions to the specific agent handling that listing type in that neighborhood.

Our approach

1. 1

   MLS as a live source of truth

   Rather than snapshotting listings, the system queries live MLS on every relevant turn. Prices change, listings pull, days-on-market ticks — the AI has to reflect what's true right now or it introduces liability.

2. 2

   Per-listing knowledge context

   Each active listing gets a retrieval-augmented context bundle — school data, HOA rules, tax details, neighborhood facts. When a caller mentions the listing, that context loads into the model's working memory before qualification even starts.

3. 3

   Agent specialization routing

   The routing layer knows which agent handles which price band and neighborhood. Booked never routes an investment buyer to an agent who only shows first-time-buyer inventory — that's a signal that trains callers to trust the system.

4. 4

   SMS confirmation loop

   Every booking sends an SMS confirmation with the calendar invite. Voice AI can misspell addresses; SMS gives the caller a chance to catch it before the appointment. This design decision cut no-shows meaningfully.

Key architectural decisions

Results

• 24/7 call coverage — zero missed listing inquiries during pilot deployments
• ~40% reduction in agent phone time, redirected to closings and showings
• Integrated with major CRMs (Follow Up Boss, kvCORE, Sierra Interactive) and calendar platforms
• Deployed across 3 brokerages within the first 6 months of launch
• Buyer-agent match rate improved via specialization routing
• SMS confirmation loop measurably reduced viewing no-shows
• Multi-market rollout capability — each new brokerage onboards in days, not weeks

Impact

Booked is the pattern we replicate every time we vertical-launch CAIDEN into a new industry: keep the platform, deeply learn the vertical's language and workflows, integrate the systems agents actually use, and design human handoffs that respect the professional on the other end. The brokerages using Booked don't describe it as "adding AI" — they describe it as "not losing after-hours leads anymore," which is the outcome that gets renewed.

Tech stack