AI chatbot for a fashion brand
Messenger questions were answered slowly by hand.
- Client
- Fashion e-commerce brand
- Service
- AI automation & agents
- Status
- Live
01 / Build
What I built
Messenger chatbot built on n8n with an LLM agent.
- n8n
- LLM agent
02 / Decisions
Key decisions
- Build it on n8n, so the brand's team can see and adjust each step of the workflow instead of depending on code only a developer can change.
- Start with Messenger only, because that's where the questions were arriving; another channel is a new trigger on the same workflow, not a new bot.
03 / Notes
Architecture and hindsight
Architecture
A Messenger webhook starts an n8n workflow for each incoming message. The workflow passes the message and the conversation so far to an LLM agent node, which writes the answer, and the reply goes back through the Messenger API. Because every step is a visible node, changing what the bot says or adding a step doesn't mean touching code.
What I'd do differently
Add a clear route to a person and a log of questions the bot couldn't answer from the first day. The log shows which answers are missing, and the route means a customer is never stuck with the bot.
04 / Outcome
Outcome
Live in production; Messenger questions are answered without waiting for someone to reply by hand.
Have a problem like this one?