For the last few weeks I’ve been building an “AI Moderator” for Seller Bari, my clothing store (~114 products). Customers message on WhatsApp, and the bot handles the whole conversation. A human only steps in when it’s actually needed. What it does right now: ✅ Answers product questions with live price, sizes, colors and stock (synced from WooCommerce every 30 min) ✅ Sends real product photos when a customer asks “can I see some pictures?” ✅ Takes the full order (multiple products, size/color, name, phone, address), shows a summary with the correct delivery charge for each zone, and asks for confirmation ✅ On “yes”, it creates the order directly in WooCommerce (correct variations, so stock updates too) and sends the customer their order number ✅ Looks up existing orders, including website orders, by order number, phone or email, and only shows details if the phone number matches ✅ Hands off to a human with a private note (order summary, or why it needs help) when someone wants a discount, has a complaint, or asks something it doesn’t know Stack: self-hosted n8n + Chatwoot on one VPS, Gemini (with a fallback model), Postgres chat memory, WooCommerce REST API, WhatsApp Cloud API. A few lessons if you’re learning n8n: 1️⃣ An empty “Key” in the Postgres Chat Memory node made my bot send blank messages. The AI Agent failed silently into the success branch. Always guard your reply with an “is the output empty?” check. 2️⃣ The free Gemini tier allows ~15 requests/minute per model. If your fallback is the same model, both fail together. Use a different model. 3️⃣ Don’t fetch your whole catalog on every message. Syncing products into an n8n Data Table took replies from ~8s down to ~1.5s. 4️⃣ Let the AI output simple tags ([ORDER], [IMAGES], [HANDOFF]) and let n8n do the real work. It’s much more reliable than letting the AI call APIs on its own. If you’re a business owner who gets the same questions all day on WhatsApp, Messenger or your website (and you use WooCommerce or GoHighLevel), DM me. I can build this for your business.