AI Sales Agent for Multi-Channel Lead Qualification
A conversational agent that handles inbound enquiries across Instagram, Facebook and WhatsApp, qualifies the lead, pushes it into the CRM, and books the appointment with a consultant.
600+
Conversations handled
< 30s
First response time
0
Missed overnight enquiries
3
Channels, one pipeline
The problem
Every inbound enquiry arrived in one of three inboxes and was answered by hand.
- Consultants spent the first hour of each day triaging messages that mostly went nowhere.
- Anything arriving overnight sat unanswered until morning, by which point the lead had usually gone elsewhere.
- No single view of who had been contacted, so leads were double-messaged or missed entirely.
What I built
A single agent sits behind all three channels. It answers immediately, works through the qualification questions in natural conversation, and — when the lead qualifies — offers real appointment slots and confirms the booking without anyone stepping in. Anything ambiguous is handed to a person with the conversation already summarised.
Demo
A run through the system end to end.
Before and after
Before
- Three inboxes checked manually
- First reply within hours, if at all
- Consultants triaging unqualified leads
After
- One conversation pipeline
- Replies in seconds, around the clock
- Consultants only see booked appointments
“The consultants only speak to leads who have already booked. The first-touch work just disappeared.”
Business impact
First-touch handling disappeared as a task.
- Consultants open their calendar rather than their inbox.
- Every booked appointment arrives with a qualification summary and full transcript.
- Enquiries are answered in seconds regardless of the hour.
- Unqualified conversations never reach a consultant.
Scale & reliability
Mar 2025
Live since
~2k/mo
Executions
99.2%
Success rate
~40
Concurrent conversations
Timeline & role
Client
Consultancy Group
System architecture
How it works
- 1
Single inbound webhook normalises messages from Instagram DMs, Facebook Messenger and WhatsApp into one shared conversation format.
- 2
An LLM agent runs the qualification script, asking for budget, timeline and service need in natural conversation rather than a rigid form.
- 3
Qualified leads are written straight into the CRM with the full transcript and a structured summary attached.
- 4
The agent negotiates a slot against live consultant availability and confirms the booking inside the same chat thread.
- 5
Unqualified or ambiguous conversations escalate to a human with the context already summarised.
Key decisions
One normalised pipeline over three separate workflows
Three near-identical flows would have tripled the maintenance for every prompt change. Normalising at the edge means the qualification logic exists once.
Conversational qualification over a form
Forms convert badly in a DM. Asking the same questions conversationally keeps the lead engaged and yields better answers than a link to a typeform.
Challenges & failure modes
Platform rate limits during bursts
Outbound messages queue and drain at a fixed rate, so a burst of enquiries slows rather than dropping replies.
Duplicate webhooks
Meta re-delivers on timeout. Each message is keyed and deduplicated before reaching the agent, so a retried delivery cannot produce a second reply.
Tech stack
Documentation
Full development report — scope, data model, and the decisions taken along the way.
Open documentationWant to see it running?
I demonstrate the full workflow live on Google Meet — the canvas, the logic, and the system handling real input. Happy to answer anything the write-up didn't cover.
I don't share workflow JSON. These are custom builds and the architecture is the work.
