An AI Sales Assistant That Runs on WhatsApp*
A real estate agent needed every inbound WhatsApp lead answered, qualified, and followed up on without manual work. We built the system that does it all: responds in under 2 minutes, qualifies buyers, books viewings, transcribes calls, and dispatches documents. One WhatsApp number, zero manual follow-ups.
Business Impact
Faster engagement and automated follow-ups close deals that used to slip away
Business Impact
Executive Outcomes
Leads engaged before they move to a competitor
Speed-to-lead drives significantly higher booking rates
Qualification, reminders, and bookings run without manual work
Faster engagement and automated follow-ups close deals that used to slip away
The Challenge
“The agent was losing momentum between inbound messages, CRM updates, booking coordination, document sharing, and manual follow-ups. The initial setup relied on n8n flows and spreadsheet-based lead management, creating scaling and consistency limitations.”
Leads sat unanswered for hours while the agent juggled WhatsApp, CRM, and spreadsheets manually
Follow-ups depended on memory. No system tracked who needed a callback or a document
Call recordings piled up in Google Drive with no transcription, no summaries, no CRM sync
Booking updates from Cal.com required manual copy-paste into Close.com notes
Voice outreach campaigns were impossible without a dedicated team or call center
The Transformation
What changed after we built the system
Leads sat unanswered for hours while the agent juggled WhatsApp, CRM, and spreadsheets manually
AI responds to every inbound WhatsApp message within the batching window, qualifying leads automatically
Follow-ups depended on memory. No system tracked who needed a callback or a document
48-hour follow-up reminders fire on schedule with zero manual tracking
Call recordings piled up in Google Drive with no transcription, no summaries, no CRM sync
Call recordings are ingested from Drive, transcribed with Whisper, summarized, and logged to CRM
Booking updates from Cal.com required manual copy-paste into Close.com notes
Booking lifecycle events sync directly into Close.com leads with structured notes
Voice outreach campaigns were impossible without a dedicated team or call center
Voice campaigns generate personalized audio with ElevenLabs and dispatch via WhatsApp at scale
The batching window changed everything
Real estate leads type fast. They send three messages in 30 seconds: a greeting, a question about price, and a photo of a listing they saw online.
Without batching, the AI would reply to each message individually, creating a disjointed conversation that felt robotic. The 2-minute batching window consolidates all rapid messages into a single context, producing one coherent response.
The 5-minute ceiling prevents messages from waiting too long. If no new message arrives within 2 minutes, the batch closes and the AI responds. This creates a natural rhythm that mirrors how a human agent would actually read and reply.
How We Built It
Technical architecture for the curious
Entry
Inbound messages hit the webhook, get validated, and route to admin or lead processing paths.
Orchestration
Background tasks handle batching, qualification, follow-ups, call ingestion, and campaign execution.
Data
Supabase stores lead state and conversation history. Sheets handle inventory lookups and campaign queues.
AI
AI responses use structured output for qualification scoring. Langfuse traces every model call for debugging.
Media
Voice notes transcribed on ingest. Campaign audio generated with ElevenLabs and served from Supabase Storage URLs.
Engineering Decisions
Tradeoffs we made and why
2-minute batching with 5-minute ceiling
Benefit
Coherent, human-like AI responses instead of fragmented replies to each message
Cost
Delayed first response in fast back-and-forth conversations
Fire-and-forget for non-critical paths
Benefit
Faster user-facing response times for the primary chat flow
Cost
Eventual consistency for side effects like CRM notes and welcome media
Acknowledge images without vision inference
Benefit
Significant cost savings and pipeline stability
Cost
No semantic understanding of shared images or screenshots
Supabase migration from Google Sheets for lead state
Benefit
Transactional safety, RLS policies, and proper relational queries
Cost
Higher initial engineering effort and migration complexity
Zod validation at every integration boundary
Benefit
Runtime type safety across 10+ external APIs
Cost
More upfront schema work and maintenance overhead per integration
*The Meta-verified WhatsApp Business API account and phone number were provided by the client. BrownMind built the AI system that operates on top of it. We do not handle WhatsApp account registration, Meta Business verification, or number provisioning.
Certain client names, proprietary workflows, screenshots, and internal assets referenced in this case study are protected under a non-disclosure agreement and have been anonymized or omitted to comply with our confidentiality obligations.
Want a system like this for your team?
Book a free 30-minute call. We will walk through your current workflow, find the biggest time sinks, and tell you exactly what we would automate.
30 minutes with Apurva. Not a sales call.
Book Your Free Audit