Fallback Answer Template

Outcome

When the agent cannot answer well, the customer gets a helpful, consistent response and a path forward instead of a made-up reply.

Main risk

Fallbacks feel robotic or trap the customer in a loop with no real way out.

Prevention

Keep fallbacks short and varied, always attach a next step, and cap repeats before forcing an escalation.

Fallback

After two unresolved fallbacks in a session, hand off to a human or capture the request for follow-up.

Architecture overview

Fallback Answer Template uses a bounded agent handoff layer for AI Agents. The designed 'I don't know' path: a small library of safe, branded responses the agent uses when it lacks a grounded answer, hits a blocked topic,... The architecture connects enumerate the failure cases, openai, vapi, and agent handoff with an explicit control path.

• Conversation layer: Enumerate the failure cases (no retrieval match, blocked topic, tool error, low confidence, off-scope)
• Reasoning layer: Write one concise on-brand fallback per case, each ending in a concrete next action (book, capture, escalate)
• Tools layer: OpenAI runs the bounded conversation step for Fallback Answer Template while keeping tool use, transcripts, and escalation outcomes explicit.
• Records layer: Vapi connects calls, messages, calendar work, or CRM writes while keep fallbacks short and varied, always attach a next step, and cap repeats before forcing an escalation.
• Escalation layer: When the agent cannot answer well, the customer gets a helpful, consistent response and a path forward instead of a made-up reply.

Data flow

1. Enumerate the failure cases (no retrieval match, blocked topic, tool error, low confidence, off-scope)
2. Write one concise on-brand fallback per case, each ending in a concrete next action (book, capture, escalate)
3. Wire the agent to select the matching fallback instead of free-generating when a guardrail fires
4. Log which fallback fired and how often, to find gaps worth turning into real answers

Controls and fallbacks

• Fallbacks feel robotic or trap the customer in a loop with no real way out.
• Keep fallbacks short and varied, always attach a next step, and cap repeats before forcing an escalation.
• After two unresolved fallbacks in a session, hand off to a human or capture the request for follow-up.

Tools

• OpenAI
• Vapi
• Retell
• GoHighLevel