ValueEdge Property Group (Australia)
AI-Powered Outbound Voice Agent System for Property Valuation Campaigns
After 4 months of development and initial campaign runs, here are the real numbers:
Case Study: AI-Powered Outbound Voice Agent System for Property Valuation Campaigns
Client Overview
Client: ValueEdge Property Group (Australia) Industry: Property Valuation & Real Estate Advisory Location: Australia-wide operations Project Duration: 4 months Built by: Bitontree (Bhavin Karad & Yash Vibhandik)
The Problem
ValueEdge Property Group runs property valuation services across Australia. Their sales and marketing teams were manually calling hundreds of leads every week to promote free property valuation guides (ebooks), book consultation meetings, and move prospects through their pipeline.
The problems were obvious:
- Sales reps spent 70% of their time dialing and leaving voicemails instead of closing deals
- No consistent follow-up. Leads fell through cracks between campaigns
- Zero visibility into what was working. No one could answer basic questions like "how many people actually picked up?" or "what's the sentiment during calls?"
- Campaign setup was a mess. Every new marketing push required manual coordination between teams, spreadsheets, and calendar tools
- Ebook distribution was manual. After a call, someone had to remember to email or WhatsApp the guide
They needed a system that could run outbound call campaigns autonomously, deliver value to prospects (the ebook), capture lead info, book calendar meetings, and report back on everything, without a human touching the phone.
The Solution
We built a full-stack AI outbound calling platform that automates the entire campaign lifecycle, from lead upload to booked meeting to analytics.
The system has five core modules:
1. Campaign Configuration Engine
A React-based dashboard where ValueEdge's marketing and sales teams can create and configure campaigns independently.
Each campaign is fully customizable:
- Campaign type (marketing nurture, sales outreach, re-engagement, event promotion)
- AI agent persona and script (tone, talking points, objection handling)
- Ebook/resource to deliver (property valuation guides, market reports, suburb-specific analyses)
- Target audience filters
- Call scheduling rules (days, time windows, timezone handling for AU states)
- Auto-trigger settings (launch automatically at configured date/time or manual trigger)
No developer needed. The marketing team spins up a new campaign in under 10 minutes.
2. Lead Management System
Three ways to get leads into the system:
- Bulk upload via CSV/Excel (validated on import, duplicates flagged automatically)
- Single lead entry via form (for walk-ins, referrals, or manual adds)
- API endpoint for CRM integration (future-ready)
Each lead record tracks: name, phone, email, property address, source, campaign assignment, call history, and current status.
The system handles deduplication, validates phone numbers for Australian format, and auto-assigns leads to their configured campaign.
3. AI Voice Agent (The Core Engine)
This is where the real magic sits.
When a campaign triggers, the system picks up leads in queue and initiates outbound calls through Twilio. The voice interaction is powered by Vapi/Retell AI for real-time voice orchestration, ElevenLabs for natural-sounding Australian English speech, and LangChain for conversation management and dynamic responses.
The call flow:
Step 1: Introduction. The agent introduces itself and ValueEdge, referencing the specific campaign context ("Hi, I'm calling from ValueEdge Property Group. We've put together a free property valuation guide for homeowners in [suburb]...")
Step 2: Value delivery. The agent explains the ebook's key insights based on the campaign configuration. Different campaigns load different talking points, so the same system handles a first-home-buyer guide differently from an investment property market report.
Step 3: Lead qualification. The agent collects or confirms: full name, email address, and best contact number.
Step 4: Calendar booking. If the prospect shows interest, the agent checks real-time availability on Google Calendar and books a consultation meeting on the spot. The prospect gets an instant calendar invite.
Step 5: Ebook delivery. Post-call, the system automatically sends the relevant ebook via email or WhatsApp based on the prospect's preference. This runs through automated workflows, no human in the loop.
The agent handles objections, answers basic questions about the valuation service, and knows when to gracefully end a call if the prospect isn't interested.
4. Post-Call Automation Layer
After every call, several automations kick in:
- Ebook delivery via email (SendGrid) or WhatsApp (Twilio) based on prospect preference
- Calendar invite sent to both the prospect and the assigned ValueEdge consultant
- Lead status updated in the system (called, interested, meeting booked, not interested, callback requested)
- Call recording and transcript stored for quality review
- Sentiment score tagged on the lead record
5. Analytics & Reporting Dashboard
Real-time analytics dashboard built in React that answers every question the sales and marketing teams were asking:
Campaign Performance Metrics:
- Total calls made vs. calls picked up (pickup rate)
- Ebook delivery rate (email vs. WhatsApp breakdown)
- Calendar meetings booked per campaign
- Cost per call and cost per booked meeting
Call Quality Metrics:
- Average call duration
- Sentiment analysis per call (positive, neutral, negative)
- Common objections and drop-off points
- Conversion funnel: called > picked up > engaged > ebook sent > meeting booked
Lead Intelligence:
- Lead status distribution across campaigns
- Best-performing time slots and days
- Geographic performance (which AU regions convert better)
- Campaign comparison (A/B performance across different ebook types or scripts)
Tech Stack
| Layer | Technology |
|---|---|
| Frontend | ReactJS, TailwindCSS |
| Backend | Python (FastAPI) |
| Database | MongoDB |
| AI/LLM | LangChain, OpenAI GPT |
| Voice AI | Vapi / Retell AI |
| Voice Synthesis | ElevenLabs (Australian English) |
| Telephony | Twilio (outbound calls, WhatsApp) |
| Calendar | Google Calendar API |
| SendGrid | |
| Automation | Custom Python workers, cron-based campaign triggers |
Architecture Overview
[Campaign Dashboard (React)]
|
v
[Campaign Config + Lead Upload API (FastAPI)]
|
v
[MongoDB - Campaigns, Leads, Call Logs, Analytics]
|
v
[Campaign Scheduler / Auto-Trigger (Python Worker)]
|
v
[Twilio Outbound Call] --> [Vapi/Retell AI Voice Orchestration]
| |
| [ElevenLabs TTS]
| |
| [LangChain Conversation Manager]
| |
| [Google Calendar API - Book Meeting]
|
v
[Post-Call Automations]
| | |
[Email] [WhatsApp] [Analytics Pipeline]
| | |
v v v
[SendGrid] [Twilio] [Sentiment Analysis + Dashboard]
Results
After 4 months of development and initial campaign runs, here are the real numbers:
| Metric | Result |
|---|---|
| Call Pickup Rate | 25% |
| Meeting Booked (of pickups) | 6% |
| Ebook Delivered (of pickups) | 50% |
| Cost Per Minute | ~$0.15/min |
| Campaign Setup Time | Under 10 minutes (was 2+ hours manual) |
| Manual Calling Effort | Reduced to near zero |
| Lead Follow-up Gap | Eliminated (100% automated) |
To put this in perspective for a typical campaign of 1,000 leads:
- 250 calls picked up
- 125 ebooks delivered automatically
- 15 consultation meetings booked on calendar
- Zero manual phone calls made by the sales team
The cost per booked meeting came in significantly lower than their previous model of salaried sales reps making manual calls, where pickup rates were similar but follow-up ebook delivery was inconsistent and meeting booking required back-and-forth over multiple days.
Before vs. After
| Process | Before | After |
|---|---|---|
| Campaign setup | 2+ hours, multiple spreadsheets, manual coordination | 10 minutes in dashboard |
| Lead upload | Manual entry or emailed spreadsheets | Bulk upload, Excel import, or form entry |
| Outbound calling | Sales reps manually dialing all day | AI agent calls autonomously on schedule |
| Ebook delivery | Manual email after call (often forgotten) | Automatic via email or WhatsApp within seconds |
| Meeting booking | Back-and-forth phone/email over days | Booked on the call, instant calendar invite |
| Analytics | "I think we're doing okay?" | Real-time dashboard with sentiment, conversion, and cost data |
| Follow-up consistency | Inconsistent, leads lost between handoffs | 100% of leads contacted, tracked, and followed up |
Key Technical Challenges We Solved
1. Australian Timezone Handling
Australia has multiple timezones (AEST, ACST, AWST) and daylight saving variations. We built timezone-aware scheduling that ensures calls only go out during appropriate business hours for each lead's location.
2. Natural-Sounding Conversations
Early iterations sounded robotic. We fine-tuned ElevenLabs voices for Australian English, adjusted pacing and pauses in the conversation flow, and built LangChain prompt chains that handle interruptions and tangential questions naturally.
3. Real-Time Calendar Availability
The agent checks Google Calendar availability during the live call and offers specific time slots. This required sub-second API responses to avoid awkward pauses. We implemented calendar caching with 5-minute refresh cycles to keep the conversation flowing.
4. Sentiment Analysis at Scale
We built a lightweight sentiment classifier that runs on every call transcript post-call. This feeds the analytics dashboard and helps ValueEdge identify which campaign scripts and ebook topics generate the most positive engagement.
5. Campaign Isolation
Multiple campaigns can run simultaneously without interference. Each campaign has its own lead queue, script configuration, scheduling rules, and analytics. One campaign's performance doesn't affect another's execution.
What Made This System Different
This isn't just a voice bot that reads a script. It's a full campaign operations platform:
- Self-service campaign creation. Non-technical marketing staff configure everything
- Intelligent conversation. The AI adapts based on prospect responses, not a rigid decision tree
- End-to-end automation. From the first ring to the calendar invite to the ebook in their inbox, no human touches the process
- Actionable analytics. Not vanity metrics. Real data on what's converting and what's not, down to sentiment per call
The system replaced a process that previously required 3-4 sales reps making calls full-time, and it runs 24/7 within configured time windows without burnout, sick days, or inconsistent messaging.
Team
Bhavin Karad - AI/Voice Agent Architecture, LangChain Integration, Twilio/Vapi Setup, Campaign Engine, Analytics Pipeline
Yash Vibhandik - Full-Stack Development, React Dashboard, API Layer, Lead Management System, MongoDB Design, Post-Call Automations
Built by Bitontree - AI-first development team specializing in voice agents, automation, and SaaS platforms.
Timeline
| Month | Deliverable |
|---|---|
| Month 1 | Campaign configuration engine, lead management system, database architecture |
| Month 2 | AI voice agent core (Twilio + Vapi/Retell + ElevenLabs + LangChain), call flow design |
| Month 3 | Google Calendar integration, post-call automations (email + WhatsApp), sentiment analysis |
| Month 4 | Analytics dashboard, multi-campaign support, testing, production deployment |
Built with Python, ReactJS, MongoDB, Twilio, Vapi/Retell AI, ElevenLabs, LangChain, and Google Calendar API. Project duration: 4 months | Bitontree (Bhavin Karad & Yash Vibhandik)