The Rundown: In this guide you'll learn how to build a time-saving AI customer support system using OpenAI's Agent Builder that automatically routes inquiries, provides intelligent responses from your documentation, and integrates directly with your website.

Who is this useful for:

• Support teams handling repetitive questions across multiple topics
• Product managers testing AI support flows before involving developers
• Startups building customer service without hiring a full support team
• Anyone needing smart routing between different support categories
  

STEP 1: Access Agent Builder and add credits

Head to Agent Builder Platform and sign into your OpenAI account. Before building anything, go to the Billing section in settings and add credits to your account - OpenAI deducts from this balance based on your usage.

Once you have credits loaded, you're ready to start building your support workflow.

STEP 2: Create your workflow and set up the routing agent

Click the "+ Create" button to start a new workflow. Every workflow begins with a start node - leave the defaults as they are.

Add your first agent node and configure it as a routing agent with a prompt that could be like:

"You are a customer support classifier. Classify the user's intent into one of the following categories: 'product_info' and 'billing_info'. Any questions about features, specifications, or how to use products should route to 'product_info'. Any payment issues, subscription questions, refunds, or pricing inquiries should route to 'billing_info'.

This agent reads incoming customer questions and decides which specialized agent should handle them.

Additionally, configure the routing agent's output format to be structured JSON by setting a response schema:

• Name the property "classification" to hold the routing decision
• Set the property type to "enum" from the dropdown menu
• Add the property values as "product_info" and "billing_info"
• Click "Update" to save the configuration

STEP 3: Create the conditions and configure the specialized agents

Connect your routing agent to a conditional block (if-else node). Where you configure it by entering the conditions, In our case the conditions to be entered would be:
input.output_parsed.classification == "product_info” and

input.output_parsed.classification == "billing_info”

Now create two specialized agent nodes with specific instructions:

1. Billing Agent:

"You are a billing support specialist. Help customers with payment issues, subscription questions, and refund requests. Be clear and provide specific next steps based on the policies mentioned in the 'Billing Policies' doc"

1. Product Info Agent:

"You are a product specialist. Answer questions about features, how-tos, and product capabilities. Use the ‘Product info’ documentation file to provide accurate answers."

For both the agents upload configure the “File Search” tool by simply uploading the relevant company docs and then connect both these agents to the conditional branches from the previous node.

STEP 4: Test and publish your workflow

Click the preview button to test your workflow. Try different customer scenarios:

• "How do I use the analytics feature?" should route to product info
• "I was charged twice this month" should route to billing

Watch the flow in preview mode - you'll see the message hit the routing agent, get classified, then route to the correct specialized agent with the appropriate response.

When everything works correctly, click Publish and name your workflow something like "Customer Support AI." You'll get a workflow ID that you can integrate into your application using Chatkit.

Pro tip: You can also repurpose this routing pattern for sales qualification, lead scoring, or internal help desk systems - the same structure works anywhere you need intelligent categorization and specialized responses while using Widget Studio to design custom response formats like order status cards.