Fallback Response
A pre-defined AI response triggered when the caller's intent cannot be understood.
A fallback response is the pre-defined behavior an AI voice agent uses when it cannot confidently understand or fulfill a caller's request. Rather than guessing or going silent, the agent follows a graceful recovery path. Fallback design is what keeps a conversation from collapsing when the model hits an edge it was not built for.
Common fallback strategies
- Reprompt / clarify: "I didn't catch that — could you say it another way?" Best for transient recognition failures.
- Rephrase the question: ask in a more constrained way to narrow the caller's options.
- Offer a menu or DTMF: drop to structured input when free-form speech keeps failing.
- Escalate to a human: a warm transfer after repeated failures, so the caller is never trapped in a loop.
- Capture and follow up: take a callback number when no path resolves live.
Why it is critical
The worst caller experience is an agent that confidently does the wrong thing, or loops endlessly. A well-tuned fallback hierarchy — clarify, then simplify, then escalate — caps the number of failed attempts and guarantees every call has an exit. Fallback rates are also a key analytics signal: a spike points to a knowledge-base gap or an ASR problem worth fixing.