Summary

Cars24 deploys OpenAI-powered voice and chat agents to automate customer conversations at scale. The system handles over 1M monthly conversation minutes, recovers roughly 12% of previously lost leads, and enables agentic workflows that empower multiple teams across the company.

Key Points

• Architecture: mix of streaming voice and chat agents with orchestration for multi-step, agentic workflows.
• Scale: >1M monthly conversation minutes requiring streaming I/O, async processing, batching, and horizontal scaling.
• Business impact: ~12% recovery of lost leads through automated outreach and real-time conversational routing.
• Engineering practices: implement observability, cost controls, human-in-the-loop fallbacks, and privacy/compliance safeguards.
• Integration: use model responses for decisions + connectors to CRM and internal workflows; prefer prompt engineering, instruction tuning, or fine-tuning as needed.

Practical tips for engineers

• Prioritize streaming APIs and timeout-recovery paths to keep voice latency low.
• Use queueing and idempotent workers for durable message processing and retries.
• Instrument end-to-end metrics (volume, latency, conversion, cost per convo) and surface alerts for regressions.
• Design clear escalation rules to hand off to human agents and log context for audits.

Bottom line

Cars24 demonstrates that combining voice and chat agents with agentic orchestration can deliver measurable lead recovery and scale conversational experiences—engineers should focus on streaming, resilience, observability, and cost control when implementing similar solutions.