06 · AI Customer Support Agent
E-commerce support agent
Premium spirits D2C — AI agent that handles 70% of repetitive tickets in under thirty seconds.
- Year
- 2026
- Duration
- 14 days + ongoing
- Investment
- €2,490 setup + €390/mo
- Discipline
- AI · Support · Automation
Outcome
70%
of repetitive tickets handled by AI in under thirty seconds
─ 01 · Challenge
The challenge.
Premium spirits e-commerce, 200 SKUs across whisky, rum, gin, champagne, and cognac. The founder was answering 80+ daily emails about stock, shipping, gift wrapping, and pairing recommendations. Every email was either "is this in stock" or "what pairs with X" — high volume, low complexity, but the founder was the only one who knew the catalog deep enough to answer. They needed an AI agent that could handle 60–70% of these without sounding like a robot.
─ 02 · Work
The work.
Fourteen days. Days one to three: ingested help docs, product DB, shipping policies, and sommelier notes into a vector DB. Days four to seven: tuned Claude Opus 4.7 on brand voice (analysed 200+ past replies from the founder). Set confidence thresholds — if AI isn't 85% sure, hands off to the founder. Days eight to eleven: integrated into the Crisp chat widget and an email auto-responder. Days twelve to fourteen: soft launch to 10% of traffic, fixed edge cases, full rollout.
─ 03 · Outcome
The outcome.
Now handles 70% of tickets in under thirty seconds. The founder spot-checks the AI's answers weekly — accuracy holds. Customer satisfaction scores actually went up because response time dropped from a four-hour average to thirty seconds. The €390/mo monitoring catches drift before users see it.
Process · 4 phases
01
Knowledge ingest
Day 1–3
Imported help docs, product DB, shipping policies, returns rules. Vector embeddings. Tested queries with the founder.
02
Voice + guardrails
Day 4–7
Tuned Claude on brand voice using 200+ past replies. Set confidence thresholds. Defined human handoff triggers.
03
Integration + UI
Day 8–11
Hooked into the Crisp chat widget. Email auto-responder live. QA against real ticket samples.
04
Launch + monitoring
Day 12–14 + ongoing
Soft launch to 10% traffic. Fixed edge cases. Full rollout day fourteen. €390/mo monitoring covers drift, content updates, tuning.
What shipped
200
SKUs · 4 languages · ongoing monitoring retainer
- 0170% of tickets handled by AI in under 30 seconds
- 02Customer satisfaction scores increased (faster responses)
- 03Founder reclaimed 3 hours per day
“I was losing my mind to repetitive emails. Now the agent handles them better than I did when I was tired.”