Designing AI + Human-in-the-Loop Workflow in n8n (Dynamic Product Info via Messenger)

Help me Build my Workflow
1

Hey everyone — I need some advice before I dive into building a complex n8n workflow for a client.

:puzzle_piece: The Problem

My client runs a Facebook Page and gets tons of messages like:

BUT:

  • Only he knows the stock status and prices (they change a lot on a daily basis).
  • There are way too many products to keep track of.
  • He doesn’t have time to answer everything himself…

:white_check_mark: The Goal

I want to build an n8n flow where:

  1. A customer sends a message.
  2. The AI (e.g., GPT) checks a “memory” (Airtable, Sheets, DB) for stock & price info.
  3. AI Draft:
  4. If info found in DB: AI generates a draft response using that data.
  5. If not: AI drafts a default “checking on it” or “unavailable” message.
  6. Client Review & Feedback: message generated by AI gets sent to my client (Telegram/Discord/Email). client doesn’t edit the message directly. Instead, they give natural language feedback to the AI, like:
  • “Yes, Product X is 50 dollars and in stock.”
  • “No, Product X is out of stock, back in 2 weeks.”
  • “Looks good, send.”
  1. AI Refine & DB Update:
  • The AI interprets client’s feedback (e.g., extracts 50 dollars, in stock).
  • It then generates the final polished message for the customer.
  • This extracted structured data updates my database. So next time someone asks about “Product X,” the AI already has the latest info.

Finally: Client Approval & FB Send: Client sees the final message, approves, and it goes to the customer.

I do have experience in coding and API,credentials…ect, but I am wondering if this is doable using n8n, or should i just switch to another tool ? how would you approach this problem (tools, workflow ect)?

2

Yes, it’s feasible, although there are several critical points where you’ll need to integrate external services (e.g., OpenAI, Airtable, Telegram, Messenger API). Here’s a practical way to structure it:

Input (Webhook or Facebook Messenger) -> Fetch data from memory (Airtable Node or HTTP Request) -> AI Drafting Node -> Send to customer (human-in-the-loop) (Telegram Node, Discord, or Send Email) -> Interpret feedback -> Update database -> Generate final message + approval

This flow is entirely feasible with n8n, as long as you know how to connect external services (via API or native nodes). What you’re proposing is a hybrid intelligent system, very useful for scaling customer service without losing the human touch.

1 Like
3

Right, thanks for giving me the workflow I should follow that’s exactly what I needed, this will surely be useful in the future, I did start earlier today building out such agent with n8n and it’s awesome, didn’t know we could build such stuff with n8n really awesome.

1 Like
4

Sent you a DM. I am building a better custom HITL system. Also feel free to reach out

5

For your Human-in-the-loop step you could consider gotoHuman (verified n8n node).
You can freely define the fields/content you need to be reviewed and want to be editable. Response with approved/edited content goes back to your workflow.