Case study

A UK / EU fashion DTC brand

Anonymized. We’ll walk through the full deployment on a call.

The starting point

• ~50 customer-service cases per day, ~1,500/month — roughly 1.25 FTE of work.
• Tight margins. Return shipping ate a meaningful chunk of order value.
• Multilingual customer base (English-dominant, with Spanish, Dutch, French).
• Strong commercial reason to settle return disputes with partial refunds rather than process full returns.

What we built

• End-to-end Sophie running on a leading automation platform — Gmail trigger, Shopify lookups, AI reply, structured operational state.
• A two-ladder negotiation model — a default 15→65% partial-refund ladder, plus a separate three-rung Recovery flow for threat cases.
• Manual Takeover system — flagged senders bypass Sophie automatically, route to the human team.
• Full audit logging — every reply, every escalation, every refund offer recorded.

Outcomes

• Sophie handles roughly 80% of return requests without a return label being issued.
• ~99% scenario uptime.
• ~1.1% failure rate on processed cases.
• Operating cost: a small fraction of what a full-time CS hire would have cost.

What we keep humans on

• Lost parcels and missing-package claims that need real investigation.
• Wrong-physical-item complaints (Sophie collects evidence; humans decide remediation).
• Abuse, regulatory threats, anything outside Sophie’s defined scope.
• Pre-shipment order modifications.

We can talk through the specifics on a call — including the exact prompts, the escalation rules, and how we handle edge cases.