2020 to 2025
Knowledge First, Then AI
Services
AI deployment, knowledge rebuild, ticket routing
Client
Archetype Themes
Team
Global support team of 12
Thousands of merchants. The same questions. A team of 12.
Kapa.ai
RAG
Shopify Themes
Support Ops
Escalation Design
Overview
Archetype Themes sells premium Shopify themes to thousands of merchants. Repetitive “how do I” tickets were burying the team, and an AI bot on top of thin documentation would only have produced wrong answers faster. The fix had to start with the content, not the bot.
Step // 1
Rebuild the knowledge first
Before deploying anything, I rewrote the help center, internal docs and macros into one accurate, theme-specific source of truth. If a support agent couldn’t answer from it, the AI couldn’t either.
Step // 2
Deploy Kapa.ai on clean sources
With the knowledge in shape, Kapa.ai answered merchant questions straight from the rebuilt docs. Accurate answers, cited sources, and no ticket created.
Step // 3
Route the hard stuff to people
LLM-assisted classification sent complex theme, integration and app conflict tickets straight to specialists. Every automated answer ran inside guardrails: QA review, SLAs and clear escalation to a human.
Results
Fewer tickets. Faster answers. Better support.
The team stopped answering the same questions and spent its time on the problems that actually needed a human.
20%
Fewer low-value tickets
Repetitive questions answered by AI from the rebuilt knowledge base.
90%
CSAT
Merchant satisfaction stayed high with AI in the loop.
80%
First contact resolution
Most issues solved in a single reply.
3 hrs
First response time
Across a global team covering every time zone.
Conclusion
An AI agent is only as good as the knowledge content feeding it. Most of the deployment work happens before the AI goes live: cleaning the sources, setting the guardrails, and deciding exactly when a human takes over.
more projects