What happens when the AI can't answer a question?

It hands the conversation to your team. The customer gets a clear message that a person will take over, not a vague guess or an invented policy detail, and everything already said stays attached to the thread.

Escalation is the intended behavior, not a malfunction.

What triggers a handoff

• The customer asks for a human, an agent, or a live person.
• Frustration that's better met with empathy and real authority.
• A question that needs store-specific judgment the assistant can't verify from catalog or order data.
• A workflow outside your enabled tools or store policy: exceptions, partial refunds, custom fulfillment.

In every case the assistant's job is the same. Stop guessing, say your team will take over, keep the thread intact.

What the customer experiences

They stay in the same widget or email channel. No new ticket, no re-entering an email address, no retelling the story when the conversation already holds their order number, product names, or photos. The only thing that changes is who's replying.

What your team gets

A notification, plus the full history: what the assistant tried, suggested, or looked up. If it already verified an order number or pulled tracking, staff start informed instead of re-asking. For product disputes and policy edge cases, they see the customer's exact wording and the assistant's last reply before takeover.

A bot that loops "I didn't understand" teaches customers to stop opening the widget. A warm transfer, limit acknowledged and context attached, teaches them the widget leads somewhere.