The situation
LumaGlow is a direct-to-consumer skincare brand doing $18M in annual revenue through Shopify Plus. Their support team of 5 agents handled approximately 3,200 tickets per month, with the majority being WISMO inquiries, order status checks, and basic product questions.
Their support cost was $4.50 per ticket, fully loaded. The team was burning out. Turnover was high. And the CEO was spending time managing support operations instead of growing the business.
The challenge
LumaGlow needed to:
- Reduce L1 support costs without sacrificing customer experience
- Free their best agents from repetitive ticket work
- Maintain their 4.5+ CSAT score during any transition
- Achieve measurable results within 30 days
They had tried an AI copilot from their helpdesk vendor. It helped agents respond 20% faster but didn't reduce headcount or fundamentally change the cost structure.
What we did
Week 1: Assessment & Training
We connected Atlas to LumaGlow's Gorgias helpdesk and Shopify store. Our team analyzed 90 days of ticket history to identify:
- 42% WISMO inquiries
- 18% order modification requests
- 15% return/exchange inquiries
- 12% product questions
- 13% complex issues requiring human judgment
We trained Atlas on LumaGlow's product catalog, shipping policies, return policies, and brand voice guidelines.
Week 2: Co-Pilot Mode
Atlas processed tickets with human oversight. Every response was reviewed by a LumaGlow agent before being sent. During this phase:
- Confidence thresholds were calibrated
- Edge cases were identified and addressed
- Brand voice was fine-tuned based on agent feedback
- Escalation rules were refined
Week 3-4: Autonomous Operation
Atlas transitioned to autonomous operation for ticket types where confidence exceeded 0.90. Human-in-the-loop remained active for:
- Refunds over $75
- Returns outside the standard window
- Customer complaints with negative sentiment
- Any ticket type not yet trained
The results
After 30 days of autonomous operation:
| Metric | Before | After | Change |
|---|---|---|---|
| Autonomous resolution rate | 0% | 82% | — |
| Average resolution time | 4.2 hours | 47 seconds | -99.7% |
| Cost per ticket (L1) | $4.50 | $0.95 | -78.9% |
| CSAT score | 4.5 | 4.7 | +0.2 |
| Monthly L1 cost | $14,400 | $3,040 | -78.9% |
| Projected annual savings | — | $201,120 | — |
Key observations
CSAT improved, not just maintained. This surprised everyone. The speed of resolution — 47 seconds average — delighted customers. Negative CSAT feedback dropped to near zero on autonomously resolved tickets.
Human agents became more effective. With Atlas handling 82% of L1 volume, the support team focused on complex issues. Their CSAT on human-handled tickets improved from 4.3 to 4.6.
Escalation rate was healthy. 18% of tickets were escalated to human agents. This included complex returns, product quality complaints, and edge cases. Atlas identified these correctly and routed them with full context.
What's next
LumaGlow is now evaluating Relay (Returns & Exchanges Specialist) to handle the 15% of tickets involving return processing. Projected additional savings: $38,000/year.
The CEO reports spending 6 fewer hours per week on support operations. Two of the five support agents have been reassigned to customer success and retention roles.
Ready to reduce your support costs?
Calculate your hidden labor tax in 60 seconds.