We deployed a 24/7 Dutch AI chatbot, here's what happened to their support tickets
Customer Support AI Chatbot Deployment
Built a support chatbot for a Dutch loyalty program. The kind where customers ask "how do I save points" and "what's my balance" a hundred times a day. Every question needed a human agent to respond, no after-hours support, and when agents did answer, quality varied wildly depending on who picked up.
What we built: a RAG-powered chatbot that answers in native Dutch, pulling from a single PDF stored in Google Drive. That last part is the bit I like most. The entire knowledge base is one PDF. When content needs updating, the client edits the PDF, an n8n workflow detects the change, re-chunks it, generates new embeddings with OpenAI, and re-indexes into Pinecone. No CMS, no migration scripts, no developer needed.
Every response comes back with a confidence score and an escalation flag. If the bot isn't sure or the topic is sensitive, it hands off to a human with contact info instead of guessing. Felt important to build that in from the start.
Stack: Next.js frontend, n8n for all backend orchestration (no traditional server), GPT-4.1-mini, Pinecone, Supabase.
Full case study: https://www.ayautomate.com/case-studies/customer-support-automation
For anyone who's built something similar, how do you handle the knowledge base updates? The single-PDF approach works but I'm curious if there's something better for larger docs.