All Case Studies
E-commerceHermes (Real-Time Agent) + AI Chat Engine

Hermes deflects 68% of support tickets for a 9-figure DTC brand

A 9-figure DTC brand was drowning in support tickets during peak season. Hermes handled the repetitive 80% — maintaining CSAT above the human-handled baseline while freeing the team for VIP work.

Published March 2026
68%
ticket deflection rate

The challenge

The brand was running 18-20K support tickets per month during Q4 peak, with a team sized for 6-8K. Ticket backlogs grew to 48+ hours during peak, CSAT dropped from 92 to 74, and the brand was considering doubling the team just to survive. The tickets were repetitive — order status, returns, sizing, shipping exceptions — but previous chatbot attempts had damaged brand perception and been pulled within weeks.

Our approach

  • 01

    Scoped the top 15 ticket intents and mapped them to backend data sources (Shopify, Loop Returns, Yotpo reviews, carrier APIs)

  • 02

    Tuned Hermes brand voice with 300+ of the brand's best historical support responses — the tone training is what separated this from the previous chatbot attempt

  • 03

    Deployed on the website chat widget first, with human review of every conversation for the first two weeks

  • 04

    Integrated with Gorgias help desk — Hermes handles first touch, escalates to humans with full context and suggested next action

  • 05

    Added SMS and WhatsApp channels after website chat proved out

  • 06

    Built peak-season capacity planning into the managed infrastructure

Results

  • 68% first-touch ticket deflection during Q4 peak

  • CSAT on Hermes-handled conversations: 93 (vs. 92 human baseline)

  • Average first response: 2.1 seconds (vs. 6+ hours during peak pre-Hermes)

  • Support team size held flat — Q4 handled without headcount increase

  • VIP customer response time improved because team capacity was preserved for high-touch work

Timeline

Engagement: 8 weeks to first channel live; full multi-channel deployment by week 14. Ongoing managed service.

What we learned

  • Brand voice tuning was the unlock — customers tolerated AI support specifically because it didn't sound like AI support
  • Human-in-the-loop review for the first 2 weeks caught edge cases that would have been customer-facing disasters
  • The handoff protocol (context summary + CRM record + suggested action) mattered more than the deflection rate — it kept human agents productive on escalations

“We've tried three other chatbots and pulled all of them within a month. Hermes is the first one our brand team defended instead of complaining about. And it handled Black Friday.”

— Head of Customer Experience, E-commerce

Services used in this engagement

Want a case study like this?

Tell us about your goals. We'll send a scoped proposal within one business day — no pressure, no obligation.