If you'd like to test your AI agent before going live, there are a few ways to do that.
1. Sandbox simulator
On the Core Logic, Brand Voice, and Edit SOPs/Use Cases pages, you'll find a sandbox on the right side of the screen. Use it to run through different scenarios and see exactly how the AI would respond based on your current configuration. This is a good way to test changes as you make them.
2. Native chat widget
You can test the chat experience directly through your native chat widget: https://admin.shippedsuite.com/ai/channels/web_chat. This shows you the same experience a customer would have reaching out through live chat.
You can find the chat widget configuration here: https://admin.shippedsuite.com/ai/channels/web_chat/edit.
3. Native email support address
You can also test the email experience using your native email support address, which follows the format [brand]@agent.cavalry.ai. You can locate this address under your channels here: https://admin.shippedsuite.com/ai/channels.
About test conversations
Test conversations pull from real data and execute real actions. We recommend using internal test accounts or test orders rather than live customer data.
Test conversations also use your draft settings, so you can test changes you've made before they go live. This means you can confirm a change behaves the way you expect before publishing it.
A few things to keep in mind
It can be easy to get stuck in this internal QA phase. We don't recommend spending too much time here, since the best way to see how the AI actually handles requests is through live conversations with customers. For internal QA purposes, we'd suggest keeping testing focused on brand voice, the overall flow of the SOP, and a handful of common scenarios, rather than trying to cover every possible case.
When a conversation involves customer account or order information, the AI is required to verify the customer's identity before proceeding. You can read more about our authentication process here: https://support.shippedsuite.com/content/how-does-ai-verifies-customer-identity.
If a conversation moves into a use case that isn't enabled yet, the system will automatically escalate it to your team. For example, if a customer writes in to cancel a subscription and that use case is enabled, the AI will intercept it. If the conversation later shifts into a return request and that use case isn't enabled, it will escalate to your team at that point.
If you have additional questions or need assistance, please contact support@cavalry.ai.