When you create or update your messaging configuration for the Web Widget or mobile SDK channels, you can – and should – test the setup to experience messaging conversations from the end user's perspective.
The testing processes differ for a widget with a default messaging response that does not use an AI agent, and one with an AI agent. In this article, we'll cover both scenarios.
This article includes the following sections:
Testing a widget without an AI agent
When you're creating or editing a Web Widget or mobile channel with the default messaging response, and that channel is not using an AI agent, the preview panel displays a basic preview of how an interaction might appear to your customers. This is useful for getting a general idea about the customer experience, but doesn't provide a complete picture.
You can preview the full customer experience by launching a separate testing window. You can use the testing window at any point during the widget creation or editing process.
- Create and configure a new Web Widget, or open the widget you want to update and make your changes.
- At the top of the preview panel, click Test it now.
- Click the launcher at the bottom of the testing window and begin interacting with
the conversation, as if you were a customer.
When you're done, you can close the window.
Testing an AI agent
In this section, we’ll explain a number of ways you can test your AI agent's behavior from initiating a conversation to closing a ticket, including:
- What happens when a user types in a question
- System flows like “I didn’t get that”
- The handover experience to an agent
- How and when triggers run, including Out of Office and CSAT messaging triggers.
This topic includes the following sections:
- Testing an AI agent before publication
- Testing your AI agent in a sandbox (Growth, Professional, and Enterprise plans)
- More testing tips
Testing an AI agent before publication
You can preview a new AI agent configuration before publishing it to a live channel using the Test AI agent button. You can preview any unpublished changes to an existing AI agent as well.
You can test:
- Standard responses, such as greetings or fallback messages
- Responses that vary based on business hours, or that collect data.
- Messages that include rich media, variables, or help center articles
- Transfers to agents, including conversations with real agents. Such conversations create real tickets in Zendesk Support.
To test an AI agent before publication
- Edit the AI agent you want to update or create a new AI agent.
- On the AI agent's edit page, click Test AI agent. The testing sidebar opens.
- Use the sidebar to start a conversation with the AI agent. To reset and start a new
conversation, click the Reload icon (
).
Usage and reporting
Test conversations created using the Test AI agent button don't count toward your account's automated resolutions and don't appear in related Explore reports or dashboards.
Limitations
- Answers that contain errors
- Answers that reference content from restricted help centers (where sign in is required to see content)
- Proactive messages
- Automatic translation
Testing your AI agent in a sandbox (Growth, Professional, and Enterprise plans)
Some account plans have options for testing an AI agent without exposing it to customers by creating a testing brand or using a sandbox.
If you are on a Growth or higher plan type, you are allowed to create multiple brands. You can create a brand with a help center specifically for testing purposes.
Messaging configurations and AI agents are not automatically copied into the sandbox instance and must be manually recreated. See Using messaging in your sandbox for more information.
More testing tips
There is some functionality that we recommend testing early on:
Starting and returning to a conversation
In messaging, we store the customer's conversation history so that customers can leave a chat and return later. Because of this, the AI agent is disabled once they’ve started speaking to an agent, which can make it difficult to test your AI agent.
To see a conversation as it appears initially to a customer
- Use an incognito or private browser window to start a conversation as a customer would. This way, you can see the conversation as it appears to a customer starting one for the first time. The AI agent initiates a contact with the customer, after they’ve clicked the icon to open the widget.
When a customer returns to a conversation, even after a related ticket has been solved or closed, the AI agent does not initiate the conversation. Instead, the AI agent waits for the customer to send a message.
To see a conversation as it appears to a returning customer
- In a private or incognito window, start a conversation as a customer would. Follow the conversation through agent handoff, and have an agent change the related ticket status to solved.
- Close the private browser window.
- Open the private browser window again, making sure to use the same browser, and without clearing your cache.
- Enter a message, and note the difference in how the AI agent responds.
Testing automatic translation
Automatic translation allows agents to communicate with customers even if they are using different languages. The translation feature is enabled in the Agent Workspace. You can see how automatic translation works from both the agent and customer perspectives. See Supporting multilingual AI agents and Translating conversations in the Zendesk Agent Workspace for more information.
To test automatic translation
- Create your AI agent, as described in the sections above, ensuring autotranslation is enabled.
- Change your browser language to the language you want to preview in. You can use a plugin like the Locale Switcher extension for Chrome.
- Open your AI agent and refresh the page. Note that this won’t work if you already have a conversation with an agent open.