Project · AI agent
Lucie, a voice and chat agent connected to HubSpot
Lucie answers website visitors by voice or chat in French, Dutch and English. She qualifies the request, books a meeting when it makes sense, and after each conversation records the contact, the call, a summary and follow-up tasks in HubSpot.
The problem
Most website visitors leave without filling in a form. Those who do arrive in the CRM with little context, and someone has to call them back to understand what they need.
What we built
- Voice and chat in one panel on every service page, with suggested questions adapted to the page.
- Known visitors recognised from the HubSpot tracking cookie, without exposing their email address.
- Short qualification: context, need, timeline and decision maker, then a meeting proposal.
- After each conversation: contact created or updated, call logged, summary note, follow-up task for warm leads, ticket for existing customers' support requests.
- Transparency: Lucie says she is an AI at the start of every conversation.
Architecture
The conversation runs on ElevenLabs. At the end of each call, a signed webhook reaches a Cloudflare Worker, which checks the signature, finds or creates the contact in HubSpot and writes the records. Automated tests run on the agent every week.
Website widget→Voice and chat agent→Signed webhook→Worker · validation→HubSpot
Related
Have a project in mind?
Tell us what you need to build, connect or fix.