Klaviyo Customer Agent: Building Bi-Directional Conversational Commerce at Scale
summary
Klaviyo builds a real-time, bi-directional commerce agent that serves millions of queries across web, SMS and email by combining multi-tenant RAG pipelines with browser-like HTML scraping, deterministic heuristics for text extraction, and contextual retrieval to disambiguate similar product text. An orchestrator simplifies flows into fewer LLM calls while an LLM gateway with circuit breakers handles failover and cost tracking. Production quality comes from manual error analysis, not dashboard averages.
AI-written summary · about
chapters (12)
- 0:00Klaviyo Customer Agent intro
- 2:52Brand knowledge management
- 13:02HTML text extraction war stories
- 18:20Contextual product retrieval
- 23:15Generating consumer responses
- 27:05Agent architecture simplification
- 30:57LLM gateway for reliability
- 33:25Quality metrics for AI responses
- 37:56Quality flywheel implementation
- 42:31Conversational repair methods
- 45:38Future of customizable AI agents
- 46:41Q&A
related talks
39mHow Chatty Leads Builds AI Agents Businesses Actually Want
55mChatBots Were Yesterday, Conversational Commerce Is Today
47mBuilding sophisticated conversational AI experiences
31mSeamless Customer Service with Amazon Connect & Alexa
32mHarnessing Large Language Models for Enhanced Customer Experience
54melevating customer experience w/ conversational AI
25mBuilding the shopping journeys of the future - ChatGPT Apps, Adaptive Interfaces and more...
10mBuilding Conversational AI Agents at Scale