Sales agent and inventory updates for Motosureste Suzuki

1. 01

   Motosureste Suzuki: the agent answers sales and financing questions and filters out prospects with no purchase intent, so sales advisors stopped handling every conversation and could focus on closing sales.

2. 02

   Replaced manual inventory entry across five branches: the stock PDF sent by Suzuki is processed with an AI model that extracts the data and updates WooCommerce via its REST API.

3. 03

   To quote, the agent confirms model, color, name, phone and email, shows a draft quote and, once the customer approves it, generates the formal quote and shares the link.

4. 04

   If the customer wants financing, the agent asks about the down payment and their income, then hands the conversation to an advisor. It does the same for service, spare parts and complaints. When it hands over, it tags the contact in GoHighLevel and stops replying.

5. 05

   It understands voice notes and images by turning them into text before replying, and it groups messages the customer sends in a row so it answers only once.

Motosureste's sales advisors handled every WhatsApp conversation, including those from people with no intention to buy. On top of that, stock for the five branches was entered by hand in the online store from the stock PDF that Suzuki sends.

When a message arrives, GoHighLevel sends it to n8n and the agent replies with an OpenAI model. It looks up models, prices, colors and offers in the WooCommerce store at that moment, so it does not make them up. If the customer wants a quote, the agent asks for their details, shows a draft quote for them to confirm and generates the quote in GoHighLevel. For inventory, the stock PDF arrives over WhatsApp, an AI model extracts the quantities for each model and color, and the flow writes them per branch to WooCommerce through its REST API.

• Before writing stock, the flow checks that each row's branch quantities add up to the total printed in the PDF itself. If they do not match, that row is left untouched and reported.
• Only variants that exist in the store catalog are updated. Anything the model cannot match with confidence stays as it was.
• If the store has an inventory location the flow does not know, it stops without writing anything, because updating without it would wipe that location's stock.

This was team work at Kobler. I built the agent and the inventory flow in n8n, monitored them in production and fixed their issues. I also took part in client meetings to gather business rules and identify edge cases.

• Advisors stopped handling every conversation: the agent filters out prospects with no purchase intent, and advisors focus on closing sales.
• Stock for the five branches is no longer entered by hand.
• Every quote is recorded in GoHighLevel as a sales opportunity, with the model and color the customer chose, assigned to their advisor.