A DTC Skincare Brand Cut Support Response Time From 14 Hours to Under 10 Minutes
A growing skincare brand's two-person support team couldn't keep up with order status and return questions. An AI support agent now resolves most of them without a human touching the ticket.
Support had quietly become the brand’s biggest customer experience risk, and it wasn’t obvious from the outside until the reviews started mentioning it directly.
The volume was mostly repetitive
We pulled three months of ticket history before building anything. Roughly 65% of all tickets fell into three categories: where’s my order, when will it ship, and how do I return this. None of these required judgment, they required fast, accurate access to order data the team already had, just not fast access to it.
Scoping what the agent should and shouldn’t touch
We deliberately kept the agent’s authority narrow. It can answer questions and process returns that clearly meet policy. It cannot make exceptions, offer discretionary discounts, or handle anything where a customer expresses frustration, all of which route straight to the two-person team with the full ticket history attached so nothing gets re-explained.
Why response time was the metric that mattered
The brand’s leadership was less focused on ticket deflection for its own sake and more focused on response time, since that was what was showing up in reviews. Deflecting repetitive tickets was the mechanism, but the target metric we optimized for was time to first response.
Results
Ticket volume reaching the human team dropped 42%, and average first response time across all tickets fell from 14 hours to under 10 minutes. The two-person team now spends its time on the complaints and edge cases that actually need a person, which is also where the brand sees the most reputational risk if handled poorly.
Can the AI agent actually process a return, or does it just answer questions about the policy?
It processes straightforward returns end to end, generating the return label and updating the order system, for cases that meet the brand's stated return policy. Anything ambiguous, like a return outside the standard window, gets routed to a human with the full request already documented.
How did customers respond to talking to an AI agent instead of a person?
The brand tracked CSAT before and after launch and saw no measurable drop. Most of the ticket volume was transactional, order status and shipping questions, where speed mattered more to customers than whether a human answered.
What kinds of tickets still go to the human team?
Product quality complaints, damaged items, and anything the customer expresses frustration about get routed to a human immediately rather than handled by the agent, along with any request outside the documented policies.