Conversational AI layer built for a decades-old Manila mover — voice, SMS, and Viber — into a market where no competitor had one

The brief

A decades-old Manila moving company runs almost entirely on inbound calls and web forms — every lead goes to a human coordinator, after-hours inquiries go dark, and there is no speed-to-lead automation. The brief was to build a conversational AI agent that answers the phone, qualifies the move, books a virtual survey, and routes call data to the CRM — without a human touching it. A competitive audit confirmed no moving company in Manila had built anything like it yet.

What I built

“Lisa” — a bilingual (Taglish) AI move coordinator that runs across two channels:

Phase 1 — SMS: A CBM calculator on the website fires a Twilio SMS to the lead within five minutes of form submission. The n8n workflow normalizes lead fields (name, phone, email, move date, estimated CBM, origin, destination), logs to Google Sheets, and triggers a Respond.io notification to the sales team.

Phase 2 — Voice: An ElevenLabs Conversational AI agent connected via Twilio BYOC telephony. Lisa answers inbound calls, works through 8 qualifying questions in natural conversation order (move type, date, origin, destination, property size, special items, packing needs, storage needs), pivots to booking a free Yembo virtual survey, and handles common objections with scripted fallbacks. Post-call, a Claude model extracts 12 structured fields — lead quality, survey booked, callback requested, all move details — and routes to Google Sheets, Respond.io, Movegistics CRM, and a survey confirmation SMS.

The stack was chosen to keep TCO low: Phase 1 runs at approximately $35/mo; Phase 2 adds voice for approximately $95/mo total. ElevenLabs was selected over an intermediary platform (Retell AI was the original candidate) to eliminate a per-minute markup and unify the chat and voice layers under a single provider.

How it’s built

The conversation layer is ElevenLabs Conversational AI with a custom Lisa system prompt — Taglish-first, with scripted objection handlers, a survey pivot script, PH market adaptations (barangay-aware address collection, international FCL/LCL routing, 30-90 day bridge storage prompts), and a voicemail fallback.

Telephony is Twilio BYOC SIP trunking into ElevenLabs — a Philippines DID number pointed at the agent. SMS is a parallel Twilio channel, handled entirely in n8n. Both channels funnel into the same n8n webhook routing layer, which writes to Google Sheets, notifies the team via Respond.io, and creates leads in Movegistics. Post-call analysis runs on a Claude model with 12 typed fields including lead quality classification (hot / warm / cold / not a lead).

The voice recognition prompt includes 19 Philippines-specific boosted keywords — barangay, CBM, Yembo, BCG, Ortigas, FCL, LCL, binding quote — tuned for the market’s terminology. Language is Taglish: the agent mirrors the caller’s code-switching naturally rather than forcing formal English or Tagalog.

Deployment is documented in a phased runbook: Phase 1 (SMS live in approximately one hour), Phase 2 (voice live in approximately three hours from Phase 1), Phase 3 (ElevenLabs voice upgrade after 30 days of live call data).

Why it matters

Moving companies in Metro Manila still compete on phone response time and word of mouth. The competitive audit found no AI chatbot or voice agent among the major players — Asian Tigers, Crown, Santa Fe. Lisa goes live into a genuine first-mover window.

The more durable point is architectural: the system is not a chatbot bolted onto a website. It is a qualification engine with structured output — every call produces machine-readable lead data, a quality score, and a routing decision, all without a human coordinator. The same system is a reusable fCAIO agency asset: the workflow design, prompt architecture, and n8n templates port to any service business that qualifies leads by phone.