AI sales agent for a large multi-rooftop auto dealer group

The problem

A large multi-rooftop automotive group was receiving 10,000+ inbound leads per month across multiple channels (web forms, third-party feeds, call-tracking, SMS). The existing sales team couldn’t qualify, route, or follow up fast enough. Leads went cold within hours and the lifetime conversion was bleeding.

The approach

We designed an end-to-end AI sales agent that took ownership of the entire pre-appointment funnel: ingest, qualify by intent and inventory match, run multi-touch follow-up cadences across email and SMS, schedule appointments against live dealership availability, update the CRM with all activity, and escalate the right cases to humans on the right rooftop.

Architecture

[ Architecture sketch placeholder, replace with diagram when ready. ]

• Ingestion: Webhooks from web forms, SFTP polling from third-party feeds, call-tracking event stream.
• Intent classification: LLM-based with confidence thresholds. Routes to qualification track.
• Communication: Templated + LLM-personalized email and SMS with cadence engine.
• Scheduling: Direct integration with each rooftop’s appointment system.
• CRM: All activity written back to the dealership group’s CRM with full conversation history.
• Human escalation: Triggered by confidence drops, customer requests, or specific intent signals (trade-ins, financing questions, etc.).

Results

• 10,000+ leads/month processed autonomously, end to end.
• Reduced average response time from hours to under 60 seconds.
• Sales team freed to focus on appointments and closes, not lead triage.

Stack

LLM orchestration, Python services on AWS, dealer CRM integration, twilio for SMS, scheduling APIs per rooftop.