In November 2025, a two-person home goods store on Shopify handled 1,200 customer enquiries over the Black Friday weekend. They answered 94% of them without a human being involved. Here's exactly how they set it up.
The problem
Marcus and Diana run a store selling handmade ceramic kitchenware. In 2024, the Friday-to-Monday stretch generated 340 orders and roughly 600 customer messages — questions about shipping times, return windows, whether a specific glaze was food-safe, order status. They took turns sleeping in four-hour shifts to keep up. Response times averaged six hours. Two one-star reviews mentioned "no reply."
For 2025, they wanted something different.
The setup (three documents, two hours)
They trained a RagmyAI chatbot on three documents:
- Their FAQ page, saved as a text file. Shipping times, return policy, glaze safety certifications, care instructions. Everything they'd answered more than once, written in plain language.
- A "tricky questions" document they wrote themselves. Every question that had ever stumped them — edge cases, custom order policies, whether their products were dishwasher-safe (complicated answer: depends on the glaze). They wrote honest, detailed answers for each.
- Their returns and refunds policy PDF. Word-for-word from their Shopify settings, so the AI would quote the exact policy rather than approximate it.
Setup took about two hours, including the time to write the tricky questions document. The chatbot went live on their product pages and the contact page via script tag on the Tuesday before Black Friday.
The retraining loop
They ran a soft launch for three days before the sale. Every evening, Marcus spent 20 minutes reading conversations and retraining any answer that was wrong or incomplete. In those three days he corrected eleven answers — mostly on shipping time estimates to specific countries and a question about whether custom orders were paused during the sale period (they were).
By Friday, the chatbot was answering questions Marcus would have answered himself, in the same direct tone, with the same caveats.
Black Friday weekend: the numbers
Over the four-day period:
- 1,247 chatbot conversations
- 94% resolved without a human hand-off
- 73 conversations escalated to email (Marcus answered these Monday morning)
- Average first response: under 3 seconds
- Zero one-star reviews mentioning response time
The most common question: "When will my order arrive?" The chatbot answered this correctly every time — because they'd included their exact processing and shipping estimates in the FAQ document.
What they wished they'd done differently
Train earlier. The three-day soft launch caught most issues, but there was one failure mode they only discovered on Saturday: customers asking about a specific limited-edition glaze that wasn't mentioned anywhere in the training documents. The AI gave a plausible but wrong answer. They added a document Saturday morning and retrained. A longer soft launch would have caught this.
Set clearer escalation triggers. A few customers wanted to speak to a human regardless of whether the chatbot had answered their question correctly. Adding a more visible "email us directly" link inside the chat window earlier would have reduced frustration for this small group.
Add voice. Several customers wrote in all-caps, or with multiple exclamation marks — signs of frustration. A voice-capable chatbot might have de-escalated faster. Marcus says this is on the list for 2026.
The cost
They used the RagmyAI Pro plan at USD 4.99/month. Total cost for November: USD 4.99. In 2024, they'd estimated their own time handling the same volume of enquiries at roughly 40 hours over the weekend — time they spent this year with their families.
