Direct answer: the most costly AI voice agent failures at dealerships are wrong facts delivered confidently, appointment requests mislabeled as confirmations, silent CRM write failures, and unstaffed warm transfers — not awkward synthetic voices.
Knowledge and tool failures
Use approved retrieval sources, field-level freshness rules, and explicit uncertainty paths. Tool errors must alert staff — not produce fabricated confirmations.
Conversation failures
Test interruptions, accents, stock numbers, and intent changes. Cap clarification loops and route repeated corrections to people.
Measurement failures
High "containment" can hide callers who needed a human. Audit definitions before presenting savings.
Interruption and latency
Long pauses and overlap loops erode trust faster than a short hold. Tune with real showroom noise.
Review the full failure modes guide and compliance questions.