Skip to content
AI agent projects

CARSA · Sales · Insurance · Integrations

From the first WhatsApp message to an organized sales enquiry

An AI assistant that connects CARSA's customer enquiries with business knowledge and a structured lead record. Website contact buttons and Meta messaging ads provide entry points to WhatsApp.

Implemented · Active workflowDesign and development · Iván Zavala
Conceptual illustration of a conversation connected to processing, knowledge and structured records
Conceptual project illustration. Documentary screenshots appear below.
workflow nodes
32
WhatsApp connection
API
lead and message records
Sheets

The challenge

Customer enquiries arrive from different pages and campaigns. Each conversation needs relevant answers, continuity and an organized record so a human advisor can take the next step.

The solution

I developed an n8n workflow connected to Meta's WhatsApp Business Platform API, OpenAI, a knowledge base in Supabase and Google Sheets. Conversation handling and structured data extraction have separate responsibilities.

How the system connects

  1. Website / Meta ad
  2. WhatsApp API
  3. Validation and context
  4. AI + knowledge base
  5. Reply and Sheets records

Simplified explanation of the architecture; this diagram does not execute the agent.

n8nWhatsApp Business PlatformMeta APIGoogle SheetsOpenAISupabase

Documentary material

Real project screenshots

Real n8n editor capture: the implemented conversation and lead-management workflow.
Real website capture. The WhatsApp entry link includes page and button context in the prepared message.
Real Meta creative preview with the WhatsApp button. Campaign performance is not represented.

Developed functions

WhatsApp through Meta's API

Receives incoming messages and sends replies through WhatsApp nodes, filtering delivery callbacks and other events that should not start a conversation.

Business knowledge

Uses semantic retrieval to bring relevant business information into the conversation and constrain answers to the available catalogue and context.

Conversation memory

Maintains short conversation context so the assistant can follow the enquiry without treating every message as a fresh start.

Structured contact data

A separate AI extractor and validation logic organize the information provided by the prospect before it becomes a lead record.

Lead qualification

Separates commercial interest, service compatibility and conversation state to support human follow-up.

Google Sheets records

Looks up and updates leads, records user messages and replies, and reads configurable blocked-contact and unsupported-service lists.

Human handoff

Collects the relevant contact preferences and enquiry context for an advisor, without promising a confirmed appointment when no calendar confirmation exists.

Controlled replies and fallback

Combines deterministic rules with the conversational model, respects closed conversations and provides a fallback route if the agent fails.

Would a similar system help your business?

We can review your process and define an agent with the tools, knowledge and controls your operation needs.
Tell me about your project