Retail Customer Support Agent

An e-commerce retailer wants an AI agent to handle tier-1 customer support: order status, returns, refunds, shipping. Must integrate with the order system and escalate to humans when needed. Cost per ticket matters enormously at this volume.

• 1What is the current cost per ticket and target?
• 2What systems hold order, return, refund data (APIs available)?
• 3What is the deflection target — what % of tier-1 can the agent handle?
• 4What are the escalation triggers (refund > $X, angry customer, VIP)?
• 5What is the SLA for first response and resolution?
• 6What channels — chat, email, voice, social?
• 7What is the brand voice — formal, casual, empathetic?
• 8How will you measure customer satisfaction post-interaction?

• Cost per ticket < $0.50 target (vs $5 human-handled)
• Order/return/refund actions via existing APIs with auth
• Escalation to human for edge cases and sentiment
• Multi-channel — chat, email, social
• CSAT measurement post-interaction
• Brand voice consistency
• Refund authority capped per agent action

• Tool-calling agent with order/return/refund tools
• Sentiment detection for escalation
• Refund cap with human approval above threshold
• Per-channel tone adaptation
• CSAT survey after interaction
• Cost tracking per ticket
• Fallback to human within 30s on escalation
• Brand voice guardrail in system prompt + output classifier