The Origin Story

This is something that came into my mind after having a very bad customer experience from a gym. I was trying to cancel my membership. But apparently they had closed down their customer support email, and only had an "AI Support Chatbot" on their website.

This is where it all started. Because that AI chatbot was so incredibly bad, that it could not even answer the most basic questions! (and yes, I am still stuck with that membership)

That led me on a pursuit for the next whole week, to build the ultimate Support Chatbot powered by AI…

Problem

So the problem with most support chatbots are that they are pretty dumb, or limited in what they can do. They are also usually very expensive for companies to implement from an ai agency and so on.

I wanted to create an AI Shopping Assistant for a Shopify E-commerce store, that could access all of the store's product and inventory data. I also wanted it to be able to give some guidance to the customers:

1. Styling tips - what products match together

2. Gift recommendations - give me 5 gift ideas for my mom (uses only the Shopify products)

3. Wash guidance - how do I best wash "X" product?

4. FAQ - how to I make a return? When to I get my order?

The Solution

I built a clean and simple AI workflow using n8n. Simple as it should be, to complex and agentic ai frameworks collapse or take to long time.

This workflow:

1. Triggers by a webhook (the AI chatbot interface on the Shopify store)

2. Gives the user's input to an AI Agent

3. The AI Aagent has access to 3 tools: AI LLM Model, Shopify API, Vector database

4. The AI agent chooses the most relevant tools to call and how to use them, it follows its own instructions, but have full freedom to decide its work progress.

5. The AI agent then returns a clean structured output and sends it back, and the user sees the answer back in the chat.

Simple!

So…

What is your return policy?

Input —> AI Agent —> calls Vector database (with all FAQs, policies, company info etc.) —> gives the vector data to OpenAI LLM model —> generates a clean response —> sends it back to the user.

Do you have any jackets?

Input —> AI Agent —> Shopify API app —> searches for any products in category "Jackets" —> collects 3-5 products —> LLM to summarize data —> sends back a clean response with product names, prices, sizes, and descriptions.

Summary

What did I achieve then?

Well I created an complete AI Shopping Assistant - Nova.

That can answer nearly all customer questions and actually be helpful, instead of answering "Sorry I can't help with that". It can fetch real product data in seconds. Its adaptable to any e-commerce store, just change the vector database and the system_prompt for the agent.

Taaa da! — You now can have your own shopping assistant in your store!

If you are interested in implementing something like this for your store, send me a message! Let's book a quick call and get an quick offer.