A modular AI automation system — built on n8n, OpenAI, WhatsApp and Meta's APIs — that qualifies, replies to and routes leads for client businesses before a human ever sees the conversation.
Role
AI Automation Architect
Type
Reusable automation system
Deployments
Real Estate, Beauty, Dental
Focus
n8n · OpenAI · WhatsApp · Meta API
0
Avg. first response time
0
Inquiries auto-qualified
0
Active client deployments
Overview
Project Overview
The first version of this system was built for Expandify Real Estate, to answer investor inquiries on WhatsApp in seconds instead of hours. Once it was live, the same problem showed up everywhere else: businesses losing leads to slow replies and manual CRM entry, regardless of industry.
Rather than rebuild a one-off bot for each new client, the original system was generalized into a reusable automation core — one orchestration layer in n8n, with the qualification logic, tone and CRM destination configured per deployment instead of rewritten from scratch.
Data destinationClient CRM (HubSpot & equivalents)
My Role
Building the Core, Then Adapting It Per Client
I designed, built and maintain the entire system — the orchestration architecture, the workflow library, the AI classification layer and every channel and CRM integration it connects to.
Designed the core orchestration architecture connecting inbound channels, AI classification and CRM handoff.
Built and maintain the n8n workflow library powering every deployment.
Integrated OpenAI's API for intent classification, lead scoring and reply drafting.
Connected the WhatsApp Business API and Meta's Conversions API for messaging and ad-attribution data.
Adapted the qualification logic per client while keeping one shared orchestration core.
The Problem
The Same Bottleneck, Every Industry
Hot leads go cold in hours
The first few minutes after an inquiry are the only real window most businesses get — after that, the buyer, client or patient has usually messaged a competitor instead.
Lead handling doesn't scale with headcount
Hiring another receptionist or agent for every new campaign or location isn't a real strategy for handling more inbound volume.
CRMs are only as good as the data they're given
Sales teams were manually retyping WhatsApp conversations into the CRM, when they had time to log them at all.
No two businesses qualify leads the same way
A one-size-fits-all chatbot script couldn't handle a property investor, a salon client and a dental patient with the same set of questions.
The Solution
One Orchestration Core, Configured Per Client
Instead of a bot built from scratch for every business, the system separates what stays constant from what changes. n8n owns the orchestration — receiving messages, calling APIs and moving data — while OpenAI acts as the reasoning layer that reads each conversation and decides what should happen next.
WhatsApp and Meta's APIs are the interface patients, clients and investors already use, and the CRM is the system of record every workflow ultimately writes to. Only the qualification rules, reply tone and CRM mapping change from one deployment to the next — the orchestration core stays the same.
Architecture
What Happens Between a Message and a Reply
01
Inbound message received
A conversation starts on WhatsApp, through a Meta click-to-chat ad, or from a website form — all routed into the same entry point.
02
n8n workflow triggered
A webhook fires the relevant n8n workflow, which assembles the conversation history and client-specific context.
03
OpenAI classifies the message
The message is scored for intent, urgency and qualification tier, and a contextual reply is drafted in the client's tone.
04
Reply sent through the origin channel
The response goes back out through the same channel the conversation started on, in seconds rather than hours.
05
Lead pushed to the CRM
A structured record — classification, transcript and source — is created automatically in the client's CRM.
06
High-value conversations flagged for handoff
Anything outside the system's confidence threshold, or clearly ready to close, is flagged for a human to take over immediately.
n8n
The Orchestration Layer Everything Runs On
n8n runs every workflow in the system — receiving webhooks from WhatsApp and Meta, calling OpenAI for classification, and writing structured records into each client's CRM. Its visual workflow builder means new branches, retries and error handling can be added without touching the underlying integrations.
Every client runs from the same core template, with only the branching logic specific to their qualification rules layered on top — so a fix or improvement to the core benefits every deployment at once.
0
Active workflow nodes per deployment
0
Workflow execution success rate
0
Client-specific variants of one core template
OpenAI
The Reasoning Layer Behind Every Reply
Each inbound message is sent to OpenAI's API along with the conversation history and client-specific context, and comes back classified by intent, urgency and qualification tier, with a drafted reply in the client's tone.
Anything the model isn't confident about — an unusual request, a complaint, a question outside its scope — is flagged rather than answered, so the system fails toward a human instead of a wrong answer.
0
Inquiries auto-qualified before handoff
0
Average classification time
0
Accuracy against human-reviewed samples
WhatsApp API
The Channel Where Conversations Already Happen
WhatsApp Business API is the primary channel for every deployment, since it's where investors, clients and patients were already messaging. Template messages handle confirmations and reminders, while session-window management keeps every conversation compliant with Meta's messaging policies.
0
Average first response time
0
Channel coverage
0
Inbound conversations answered
Meta API
Connecting Ad Spend to Real Conversations
Meta's Conversions API ties click-to-WhatsApp ad campaigns directly into the same automation, so a conversation started from an ad is attributed all the way through to a qualified lead — not just a link click. That attribution is what lets ad spend be measured against consultations booked, not clicks generated.
0
Click-to-chat conversations attributed
0
Lower cost per qualified lead
0
Ad campaigns feeding the pipeline live
CRM
Every Conversation Ends Up in One Place
Once a conversation is classified, n8n writes a structured record — transcript, classification and source — directly into the client's CRM, with automatic assignment to the right person on the sales or front-desk team.
No conversation depends on someone remembering to log it, and no sales team opens a lead without already knowing what it wants.
0
Conversations logged automatically
0
Manual data entry eliminated per team
0
Single source of truth per client
Workflows
The Library Every Deployment Draws From
New lead qualification
Classifies and replies to every inbound inquiry within seconds, tagging it by intent and readiness before it reaches a human.
Missed-call follow-up
Automatically re-engages anyone who called and didn't reach someone, opening a WhatsApp conversation instead of losing the lead.
Review request automation
Triggers a follow-up message after a completed appointment or sale, turning satisfied customers into public reviews without manual follow-up.
Appointment & booking reminders
Confirms and reminds clients ahead of scheduled appointments, cutting no-shows without a staff member sending each message.
Re-engagement drip
Reopens conversations with leads that went cold, spaced out over days instead of disappearing after the first unanswered message.
Automation Examples
The Same Core, Three Different Businesses
Real estate — Expandify
An investor messages after hours about a specific listing. The system replies with pricing and availability in under a minute, tags them by budget tier, and queues them for the sales team's next morning.
Beauty salon — Miled Andos
A client books a facial over WhatsApp. Two hours after the appointment, an automated message asks for a Google review with the link pre-filled.
Cosmetic dentistry — Cosmedent
A prospective patient asks about veneer pricing through a Meta click-to-WhatsApp ad. The system answers the common pricing and recovery questions immediately and flags the conversation as high-intent for the front desk to call first.
Performance
Reliable Enough to Reply in Seconds
Replies only matter if they go out fast and consistently, so every workflow execution is monitored and alerted on, API rate limits are handled with retries and backoff, and fallback logic keeps conversations moving even if one integration is temporarily unavailable.
0
Workflow uptime
0
End-to-end response latency
0
Monitored coverage
The Results
Across All Deployments
0
Avg. first response time
0
Inquiries auto-qualified
0
Active client deployments
0
Lower cost per qualified lead
Gallery
A Look at the Build
Full screenshots are being prepared for publication as new deployments continue to roll out.
n8n workflow canvasWhatsApp automation in actionOpenAI classification outputCRM lead record
Tech Stack
Technologies
n8nOpenAI APIWhatsApp Business APIMeta Conversions APIHubSpot CRMWebhooks
Next Step
Want This Running in Your Business?
The same orchestration core can be configured for a new business in days, not months — connected to your channels, your CRM and your qualification rules.