Most service-business websites still end every conversation with a form. Fill it out, wait for a call, hope the context survives the handoff. We built the opposite for a Phoenix fence contractor: a chatbot with a face, a voice, and a lead flow that captures name, project, timeline, and callback number — then hands the whole context to the owner.
The problem
Static "fill out this form" pages lose the visitors who just want to ask one question. Forms ask for everything up front and return nothing. The visitor who only wanted to know "do you do block walls?" bounces before typing their phone number into a 9-field form.
The solution
Mark — a guided chatbot embedded on the demo site with three properties most chat widgets do not have:
1. It talks and shows a face. The avatar is a live video feed (real-time WebRTC via Simli) with dynamic voice (ElevenLabs flash v2.5). Text chat runs on the same page and the same brain, so the experience is one conversation, not three widgets.
2. It is interruptible. The microphone stays live while the bot is speaking. If the visitor starts talking mid-sentence, the bot stops, listens, and answers. Replies are capped at two sentences on screen and in voice — built for back-and-forth, not monologue.
3. It runs on the client's own database. The knowledge base, vectors, and lead context live in a database the business owns, with namespace isolation at runtime. Their data, their storage — not a vendor silo.
The guided lead flow triggers only on strong booking intent ("estimate", "quote", "book", "call me"). Mid-flow questions fall through to the live brain, which answers from the contractor's knowledge base. The captured lead lands with full context: name, problem statement, timeline, callback number.
The result, in data
From our internal benchmarks across conversational landing pages: an 87% qualification rate on conversational lead capture, and 3.2x conversion versus static forms. Build-side measurements from the same pipeline: real-time avatar sessions up to 10 minutes per connection, flash-tier voice synthesis on every reply, and an interruption (barge-in) path with a 700ms echo grace so the bot never talks over the visitor — or itself.
What is next
The same template rolls across five other vertical demos, an SMS follow-up fires within two hours of an abandoned chat, and pages shift their angle based on visit history — credibility first, ROI second, a direct callback CTA for visitors three-plus visits deep.
Try the live demo: Vulcan Fence demo — tap the mic, let him start an answer, and talk over him.
If you want this for your business, the pitch in one line: chatbots that see, speak, and remember — running on a database you own.
Top comments (0)