How I Built An AI Support Agent To Automate 80% Of Tickets

Monmon Cats

Build an AI-powered support agent that could reduce the human time spent on repetitive support tasks. The agent should be able to access company policies and product information, as well as real order data from the client's Shopify store.

We built the support automation layer as the first line of support defense for a D2C brand selling on Shopify. Customers want to reach out on a variety of channels, so we pulled in any customer communication from email, sms, social, and chat as multiple trigger points for our automation. When a conversation is created or updated, the intelligent AI layer analyzes the text to determine the type of support needed. Requests were classified as:

Policy Questions (Returns/exchange information)

Order Questions (Where is my order?)

Spam/Irrelevant Questions

Complex Questions

‍

For all categories except Complex Questions, the AI assistant responds back and forth with the customer until the ticket was closed or the question becomes too complex. For complex questions that are not answerable with provided information, AI support escalates the ticket to a human agent for intervention.

To give the AI agent added context, we connect a vector database of relevant business policy and product information for our agent to search. For order specific questions, we provide external API access to Shopify, so the agent can look up order information in real time to retrieve relevant data.

This agent is still in beta, but is expected to handle up to 80% of customer support tasks without human intervention.