Problem Definition

A shop owner was spending hours every day answering the same customer queries — product details, shipping, prices, stock availability, order status — leaving almost no time to find new clients or grow revenue.

Client stack

• Product & store info: Google Docs
• Orders & product availability: Supabase (SQL)
• Storefront: WooCommerce

Goal

Give customers accurate, human-like responses automatically — freeing the owner to focus on growth.

Solution — multi-agent routing

1. Automatically intake the customer message and extract the relevant intent.
2. Classify the query into one of four types: Store, Orders, Products, or Other.
3. Route to one of three specialized agents, each with its own tools and data source:
   • Store agent — answers from the FAQ in Google Docs.
   • Order Finder — queries the orders database for products, delivery status, and expected delivery times.
   • Product agent — checks the products database for live stock and pulls descriptions from Google Docs.
4. Respond to the customer in a natural, human-like tone.
5. If the query is classified as "Other," it escalates immediately to the owner so nothing slips and the customer still gets a timely reply.

The owner stopped being the bottleneck for routine queries — and kept a human in the loop exactly where judgment was needed.

Enhancement #1 — Per-agent model selection